Pitfalls in machine learning-based assessment of tumor-infiltrating lymphocytes in breast cancer: a report of the international immuno-oncology biomarker working group

Abstract

The clinical significance of the tumor-immune interaction in breast cancer is now established, and tumor-infiltrating lymphocytes (TILs) have emerged as predictive and prognostic biomarkers for patients with triple-negative (estrogen receptor, progesterone receptor, and HER2-negative) breast cancer and HER2-positive breast cancer. How computational assessments of TILs might complement manual TIL assessment in trial and daily practices is currently debated. Recent efforts to use machine learning (ML) to automatically evaluate TILs have shown promising results. We review state-of-the-art approaches and identify pitfalls and challenges of automated TIL evaluation by studying the root cause of ML discordances in comparison to manual TIL quantification. We categorize our findings into four main topics: (1) technical slide issues, (2) ML and image analysis aspects, (3) data challenges, and (4) validation issues. The main reason for discordant assessments is the inclusion of false-positive areas or cells identified by performance on certain tissue patterns or design choices in the computational implementation. To aid the adoption of ML for TIL assessment, we provide an in-depth discussion of ML and image analysis, including validation issues that need to be considered before reliable computational reporting of TILs can be incorporated into the trial and routine clinical management of patients with triple-negative breast cancer. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Full text

Pitfalls in machine learning-based assessment of tumor-infiltrating
lymphocytes in breast cancer: a report of the international
immuno-oncology biomarker working group
Jeppe Thagaard
1,2†
, Glenn Broeckx
3,4†
, David B Page
5
, Chowdhury Arif Jahangir
6
, Sara Verbandt
7
, Zuzana Kos
8
,
Rajarsi Gupta
9
, Reena Khiroya
10
, Khalid Abduljabbar
11
, Gabriela Acosta Haab
12
, Balazs Acs
13,14
, Guray Akturk
15
,
Jonas S Almeida
16
, Isabel Alvarado‐Cabrero
17
, Mohamed Amgad
18
, Farid Azmoudeh‐Ardalan
19
, Sunil Badve
20
,
Nurkhairul Bariyah Baharun
21
, Eva Balslev
22
, Enrique R Bellolio
23
, Vydehi Bheemaraju
24
, Kim RM Blenman
25,26
,
Luciana Botinelly Mendonça Fujimoto
27
, Najat Bouchmaa
28
, Octavio Burgues
29
, Alexandros Chardas
30
,
Maggie Chon U Cheang
31
, Francesco Ciompi
32
, Lee AD Cooper
33
, An Coosemans
34
, Germán Corredor
35
,
Anders B Dahl
1
, Flavio Luis Dantas Portela
36
, Frederik Deman
3
, Sandra Demaria
37,38
, Johan Doré Hansen
2
,
Sarah N Dudgeon
39
, Thomas Ebstrup
2
, Mahmoud Elghazawy
40,41
, Claudio Fernandez‐Martín
42
, Stephen B Fox
43
,
William M Gallagher
6
, Jennifer M Giltnane
44
, Sacha Gnjatic
45
, Paula I Gonzalez‐Ericsson
46
, Anita Grigoriadis
47,48
,
Niels Halama
49
, Matthew G Hanna
50
, Aparna Harbhajanka
51
, Steven N Hart
52
, Johan Hartman
13,14
,
Søren Hauberg
1
, Stephen Hewitt
53
, Akira I Hida
54
, Hugo M Horlings
55
, Zaheed Husain
56
, Evangelos Hytopoulos
57
,
Sheeba Irshad
58
, Emiel AM Janssen
59,60
, Mohamed Kahila
61
, Tatsuki R Kataoka
62
, Kosuke Kawaguchi
63
,
Durga Kharidehal
24
, Andrey I Khramtsov
64
, Umay Kiraz
59,60
, Pawan Kirtani
65
, Liudmila L Kodach
66
, Konstanty Korski
67
,
Anikó Kovács
68,69
, Anne‐Vibeke Laenkholm
70,71
, Corinna Lang‐Schwarz
72
, Denis Larsimont
73
, Jochen K Lennerz
74
,
Marvin Lerousseau
75,76,77
, Xiaoxian Li
78
, Amy Ly
79
, Anant Madabhushi
80
, Sai K Maley
81
,
Vidya Manur Narasimhamurthy
82
, Douglas K Marks
83
, Elizabeth S McDonald
84
, Ravi Mehrotra
85,86
, Stefan Michiels
87
,
Fayyaz ul Amir Afsar Minhas
88
, Shachi Mittal
89
, David A Moore
90
, Shamim Mushtaq
91
, Hussain Nighat
92
,
Thomas Papathomas
93,94
, Frederique Penault‐Llorca
95
, Rashindrie D Perera
96,97
, Christopher J Pinard
98,99,100,101
,
Juan Carlos Pinto‐Cardenas
102
, Giancarlo Pruneri
103,104
, Lajos Pusztai
105,106
, Arman Rahman
6
,
Nasir Mahmood Rajpoot
107
, Bernardo Leon Rapoport
108,109
, Tilman T Rau
110
, Jorge S Reis‐Filho
111
,
Joana M Ribeiro
112
, David Rimm
113,114
, Anne Roslind
22
, Anne Vincent-Salomon
115
, Manuel Salto‐Tellez
116,117
,
Joel Saltz
9
, Shahin Sayed
118
, Ely Scott
119
, Kalliopi P Siziopikou
120
, Christos Sotiriou
121,122
, Albrecht Stenzinger
123,124
,
Maher A Sughayer
125
, Daniel Sur
126
, Susan Fineberg
127,128
, Fraser Symmans
129
, Sunao Tanaka
130
, Timothy Taxter
131
,
Sabine Tejpar
7
, Jonas Teuwen
132
, E Aubrey Thompson
133
, Trine Tramm
134,135
, William T Tran
136
,
Jeroen van der Laak
137
, Paul J van Diest
138,139
, Gregory E Verghese
47,48
, Giuseppe Viale
140,141
, Michael Vieth
72
,
Noorul Wahab
142
, Thomas Walter
75,76,77
, Yannick Waumans
143
, Hannah Y Wen
50
, Wentao Yang
144
,
Yinyin Yuan
145
, Reena Md Zin
146
, Sylvia Adams
83,147
, John Bartlett
148
, Sibylle Loibl
149
, Carsten Denkert
150
,
Peter Savas
97,151
, Sherene Loi
97,151
, Roberto Salgado
3,97
and Elisabeth Specht Stovgaard
22,152
*
1
Technical University of Denmark, Kongens Lyngby, Denmark
2
Visiopharm A/S, Hørsholm, Denmark
3
Department of Pathology, GZA‐ZNA Hospitals, Antwerp, Belgium
4
Centre for Oncological Research (CORE), MIPPRO, Faculty of Medicine, Antwerp University, Antwerp, Belgium
5
Earle A Chiles Research Institute, Providence Cancer Institute, Portland, OR, USA
6
UCD School of Biomolecular and Biomedical Science, UCD Conway Institute, University College Dublin, Dublin, Ireland
7
Digestive Oncology, Department of Oncology, KU Leuven, Leuven, Belgium
8
Department of Pathology and Laboratory Medicine, BC Cancer Vancouver Centre, University of British Columbia, Vancouver, British Columbia,
Canada
9
Department of Biomedical Informatics, Stony Brook University, Stony Brook, NY, USA
10
Department of Cellular Pathology, University College Hospital London, London, UK
11
Centre for Evolution and Cancer, The Institute of Cancer Research, London, UK
12
Hospital Maria Curie, Buenos Aires, Argentina
13
Department of Oncology and Pathology, Karolinska Institutet, Stockholm, Sweden
14
Department of Clinical Pathology and Cancer Diagnostics, Karolinska University Hospital, Stockholm, Sweden
15
Translational Molecular Biomarkers, Merck & Co Inc, Rahway, NJ, USA
16
Division of Cancer Epidemiology and Genetics (DCEG), National Cancer Institute (NCI), Rockville, MD, USA
17
Oncology Hospital, Star Medica Centro, Ciudad de México, Mexico
18
Department of Pathology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA
19
Tehran University of Medical Sciences, Tehran, Iran
20
Department of Pathology and Laboratory Medicine, Emory University School of Medicine, Emory University Winship Cancer Institute, Atlanta,
GA, USA
21
The National University of Malaysia, Kuala Lumpur, Malaysia
22
Department of Pathology, Herlev and Gentofte Hospital, Herlev, Denmark
23
Departamento de Anatomía Patológica, Facultad de Medicina, Universidad de La Frontera, Temuco, Chile
24
Department of Pathology, Narayana Medical College, Nellore, India
Journal of Pathology
J Pathol 2023
Published online 23 August 2023 in Wiley Online Library
(wileyonlinelibrary.com)DOI: 10.1002/path.6155
INVITED REVIEW
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and
reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

25
Department of Internal Medicine Section of Medical Oncology and Yale Cancer Center, Yale School of Medicine, New Haven, CT, USA
26
Department of Computer Science, Yale School of Engineering and Applied Science, New Haven, CT, USA
27
Department of Pathology and Legal Medicine, Amazonas Federal University, Manaus, Brazil
28
Institute of Biological Sciences, Faculty of Medical Sciences, Mohammed VI Polytechnic University (UM6P), Ben‐Guerir, Morocco
29
Pathology Department, Hospital Cliníco Universitario de Valencia/Incliva, Valencia, Spain
30
Department of Pathobiology & Population Sciences, The Royal Veterinary College, London, UK
31
Head of Integrative Genomics Analysis in Clinical Trials, ICR‐CTSU, Division of Clinical Studies, The Institute of Cancer Research, London, UK
32
Radboud University Medical Center, Department of Pathology, Nijmegen, The Netherlands
33
Department of Pathology, Northwestern Feinberg School of Medicine, Chicago, IL, USA
34
Department of Oncology, Laboratory of Tumor Immunology and Immunotherapy, KU Leuven, Leuven, Belgium
35
Biomedical Engineering Department, Emory University, Atlanta, GA, USA
36
Hospital Universitário Getúlio Vargas, Manaus, Brazil
37
Department of Radiation Oncology, Weill Cornell Medicine, New York, NY, USA
38
Department of Pathology and Laboratory Medicine, Weill Cornell Medicine, New York, NY, USA
39
Conputational Biology and Bioinformatics, Yale University, New Haven, CT, USA
40
University of Surrey, Guildford, UK
41
Ain Shams University, Cairo, Egypt
42
Instituto Universitario de Investigación en Tecnología Centrada en el Ser Humano, HUMAN‐tech, Universitat Politècnica de València, Valencia, Spain
43
Pathology, Peter MacCallum Cancer Centre and Sir Peter MacCallum Department of Oncology, University of Melbourne, Melbourne, Victoria,
Australia
44
Genentech, San Francisco, CA, USA
45
Department of Oncological Sciences, Medicine Hem/Onc, and Pathology, Tisch Cancer Institute –Precision Immunology Institute, Icahn School of
Medicine at Mount Sinai, New York, NY, USA
46
Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
47
Cancer Bioinformatics, School of Cancer & Pharmaceutical Sciences, Faculty of Life Sciences and Medicine, King's College London, London, UK
48
The Breast Cancer Now Research Unit, School of Cancer and Pharmaceutical Sciences, Faculty of Life Sciences and Medicine, King's College London,
London, UK
49
Department of Translational Immunotherapy, German Cancer Research Center, Heidelberg, Germany
50
Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, USA
51
Case Western University, Cleveland, OH, USA
52
Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, MN, USA
53
Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA
54
Department of Pathology, Matsuyama Shimin Hospital, Matsuyama, Japan
55
Division of Pathology, Netherlands Cancer Institute (NKI), Amsterdam, The Netherlands
56
Praava Health, Dhaka, Bangladesh
57
iRhythm Technologies, San Francisco, CA, USA
58
King's College London & Guy's & St Thomas’NHS Trust, London, UK
59
Department of Pathology, Stavanger University Hospital, Stavanger, Norway
60
Department of Chemistry, Bioscience and Environmental Technology, University of Stavanger, Stavanger, Norway
61
Department of Pathology, Yale University, New Haven, CT, USA
62
Department of Pathology, Iwate Medical University, Morioka, Japan
63
Department of Breast Surgery, Kyoto University Graduate School of Medicine, Kyoto, Japan
64
Department of Pathology and Laboratory Medicine, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, USA
65
Department of Histopathology, Aakash Healthcare Super Speciality Hospital, New Delhi, India
66
Department of Pathology, Netherlands Cancer Institute –Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
67
Data, Analytics and Imaging, Product Development, F. Hoffmann‐La Roche AG, Basel, Switzerland
68
Department of Clinical Pathology, Sahlgrenska University Hospital, Gothenburg, Sweden
69
Institute of Biomedicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
70
Department of Surgical Pathology, Zealand University Hospital, Roskilde, Denmark
71
Department of Surgical Pathology, University of Copenhagen, Copenhagen, Denmark
72
Institute of Pathology, Klinikum Bayreuth GmbH, Friedrich‐Alexander‐University Erlangen‐Nuremberg, Bayreuth, Germany
73
Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium
74
Center for Integrated Diagnostics, Massachusetts General Hospital/Harvard Medical School, Boston, MA, USA
75
Centre for Computational Biology (CBIO), Mines Paris, PSL University, Paris, France
76
Institut Curie, PSL University, Paris, France
77
INSERM, Paris, France
78
Department of Pathology and Laboratory Medicine, Emory University, Atlanta, GA, USA
79
Department of Pathology, Massachusetts General Hospital, Boston, MA, USA
80
Department of Biomedical Engineering, Radiology and Imaging Sciences, Biomedical Informatics, Pathology, Georgia Institute of Technology and
Emory University, Atlanta, GA, USA
81
NRG Oncology/NSABP Foundation, Pittsburgh, PA, USA
82
Manipal Hospitals, Bangalore, India
83
Perlmutter Cancer Center, NYU Langone Health, New York, NY, USA
84
Breast Cancer Translational Research Group, University of Pennsylvania, Philadelphia, PA, USA
2 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

85
Indian Cancer Genomic Atlas, Pune, India
86
Centre for Health, Innovation and Policy Foundation, Noida, India
87
Office of Biostatistics and Epidemiology, Gustave Roussy, Oncostat U1018, Inserm, University Paris‐Saclay, Ligue Contre le Cancer labeled Team,
Villejuif, France
88
Tissue Image Analytics Centre, Warwick Cancer Research Centre, PathLAKE Consortium, Department of Computer Science, University of Warwick,
Coventry, UK
89
Department of Chemical Engineering, Department of Laboratory Medicine and Pathology, University of Washington, Seattle, WA, USA
90
CRUK Lung Cancer Centre of Excellence, UCL and Cellular Pathology Department, UCLH, London, UK
91
Department of Biochemistry, Ziauddin University, Karachi, Pakistan
92
Pathology and Laboratory Medicine, All India Institute of Medical sciences, Raipur, India
93
Institute of Metabolism and Systems Research, University of Birmingham, Birmingham, UK
94
Department of Clinical Pathology, Drammen Sykehus, Vestre Viken HF, Drammen, Norway
95
Centre Jean Perrin, Université Clermont Auvergne, INSERM, U1240 Imagerie Moléculaire et Stratégies Théranostiques, Clermont Ferrand, France
96
School of Electrical, Mechanical and Infrastructure Engineering, University of Melbourne, Melbourne, Victoria, Australia
97
Division of Cancer Research, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia
98
Radiogenomics Laboratory, Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada
99
Department of Clinical Studies, Ontario Veterinary College, University of Guelph, Guelph, Ontario, Canada
100
Department of Oncology, Lakeshore Animal Health Partners, Mississauga, Ontario, Canada
101
Centre for Advancing Responsible and Ethical Artificial Intelligence (CARE‐AI), University of Guelph, Guelph, Ontario, Canada
102
Diagnostico de Salud Animal SA, Ciudad de México, Mexico
103
Department of Pathology and Laboratory Medicine, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy
104
Faculty of Medicine and Surgery, University of Milan, Milan, Italy
105
Yale Cancer Center, Yale University, New Haven, CT, USA
106
Department of Medical Oncology, Yale School of Medicine, Yale University, New Haven, CT, USA
107
University of Warwick, Coventry, UK
108
The Medical Oncology Centre of Rosebank, Johannesburg, South Africa
109
Department of Immunology, Faculty of Health Sciences, University of Pretoria, Pretoria, South Africa
110
Institute of Pathology, University Hospital Düsseldorf and Heinrich‐Heine‐University Düsseldorf, Düsseldorf, Germany
111
Department of Pathology and Laboratory Medicine, Memorial Sloan Kettering Cancer Center, New York, NY, USA
112
Département de Médecine Oncologique, Gustave Roussy, Villejuif, France
113
Department of Pathology, Yale University School of Medicine, New Haven, CT, USA
114
Department of Medicine, Yale University School of Medicine, New Haven, CT, USA
115
Department of Diagnostic and Theranostic Medicine, Institut Curie, University Paris‐Sciences et Lettres, Paris, France
116
Integrated Pathology Unit, The Institute of Cancer Research, London, UK
117
Precision Medicine Centre, Queen's University Belfast, Belfast, UK
118
Department of Pathology, Aga Khan University, Nairobi, Kenya
119
Translational Pathology, Translational Sciences and Diagnostics/Translational Medicine/R&D, Bristol Myers Squibb, Princeton, NJ, USA
120
Department of Pathology, Section of Breast Pathology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA
121
Breast Cancer Translational Research Laboratory J.‐C. Heuson, Institut Jules Bordet, Hôpital Universitaire de Bruxelles (HUB), Université Libre de
Bruxelles (ULB), Brussels, Belgium
122
Medical Oncology Department, Institut Jules Bordet, Hôpital Universitaire de Bruxelles (HUB), Université Libre de Bruxelles (ULB), Brussels,
Belgium
123
Institute of Pathology, University Hospital Heidelberg, Heidelberg, Germany
124
Centers for Personalized Medicine (ZPM), Heidelberg, Germany
125
King Hussein Cancer Center, Amman, Jordan
126
Department of Medical Oncology, University of Medicine and Pharmacy “Iuliu Hatieganu”, Cluj‐Napoca, Romania
127
Montefiore Medical Center, Bronx, NY, USA
128
Albert Einstein College of Medicine, Bronx, NY, USA
129
University of Texas MD Anderson Cancer Center, Houston, TX, USA
130
Kyoto University, Kyoto, Japan
131
Tempus Labs, Chicago, IL, USA
132
AI for Oncology Lab, The Netherlands Cancer Institute, Amsterdam, The Netherlands
133
Mayo Clinic Florida, Jacksonville, FL, USA
134
Department of Pathology, Aarhus University Hospital, Aarhus, Denmark
135
Institute of Clinical Medicine, Aarhus University, Aarhus, Denmark
136
Department of Radiation Oncology, University of Toronto and Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada
137
Department of Pathology, Radboud University Medical Center, Nijmegen, The Netherlands
138
Department of Pathology, University Medical Center Utrecht, The Netherlands
139
Johns Hopkins Oncology Center, Baltimore, MD, USA
140
Department of Pathology, European Institute of Oncology, Milan, Italy
141
Department of Pathology, University of Milan, Milan, Italy
142
Tissue Image Analytics Centre, Department of Computer Science, University of Warwick, Coventry, UK
143
CellCarta NV, Antwerp, Belgium
144
Fudan Medical University Shanghai Cancer Center, Shanghai, PR China
Pitfalls in ML assessment of TILs 3
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

145
Department of Translational Molecular Pathology, Division of Pathology and Laboratory Medicine, The University of Texas MD Anderson Cancer
Center, Houston, TX, USA
146
Department of Pathology, Faculty of Medicine, Universiti Kebangsaan Malaysia, Kuala Lumpur, Malaysia
147
Department of Medicine, NYU Grossman School of Medicine, Manhattan, NY, USA
148
University of Edinburgh, Edinburgh, UK
149
Department of Medicine and Research, German Breast Group, Neu‐Isenburg, Germany
150
Institut für Pathologie, Philipps‐Universität Marburg und Universitätsklinikum Marburg, Marburg, Germany
151
The Sir Peter MacCallum Department of Medical Oncology, University of Melbourne, Melbourne, Victoria, Australia
152
Department of Clinical Medicine, University of Copenhagen, Copenhagen, Denmark
*Correspondence to: ES Stovgaard, Department of Pathology, Herlev and Gentofte Hospital, Herlev, Denmark. E-mail: elidsp01@regionh.dk.
†
Equal contributors.
Abstract
The clinical significance of the tumor-immune interaction in breast cancer is now established, and tumor-infiltrating
lymphocytes (TILs) have emerged as predictive and prognostic biomarkers for patients with triple-negative (estrogen
receptor, progesterone receptor, and HER2-negative) breast cancer and HER2-positive breast cancer. How compu-
tational assessments of TILs might complement manual TIL assessment in trial and daily practices is currently
debated. Recent efforts to use machine learning (ML) to automatically evaluate TILs have shown promising results.
We review state-of-the-art approaches and identify pitfalls and challenges of automated TIL evaluation by studying
the root cause of ML discordances in comparison to manual TIL quantification. We categorize our findings into four
main topics: (1) technical slide issues, (2) ML and image analysis aspects, (3) data challenges, and (4) validation
issues. The main reason for discordant assessments is the inclusion of false-positive areas or cells identified by
performance on certain tissue patterns or design choices in the computational implementation. To aid the adoption
of ML for TIL assessment, we provide an in-depth discussion of ML and image analysis, including validation issues
that need to be considered before reliable computational reporting of TILs can be incorporated into the trial and
routine clinical management of patients with triple-negative breast cancer.
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great
Britain and Ireland.
Keywords: deep learning; machine learning; digital pathology; guidelines; image analysis; pitfalls; prognostic biomarker; triple-negative
breast cancer; tumor-infiltrating lymphocytes
Received 21 April 2023; Accepted 7 June 2023
Conflict of interest statement: JT: Employee of Visiopharm A/S. GB: Speaker's fee received from MSD, Novartis, advisory boards for Roche and MSD,
Consultant for MSD, Novartis and Roche, travel and conference support from Roche, MSD, and Gilead. ZK: Paid advisory role for Eli Lilly and
AstraZeneca Canada. GA: Employee of Merck. KRB: Scientific advisory board for CDI Labs, Research funding from Carevive. FC: Chair of the Scientific
and Medical Advisory Board of TRIBVN Healthcare, France, and received advisory board fees from TRIBVN Healthcare, France in the last 5 years. He is
shareholder of Aiosyn BV, the Netherlands. LAC: Participation in Tempus Algorithm Advisors program. AC: Contracted researcher for Oncoinvent AS and
Novocure and a consultant for Sotio a.s. and Epics Therapeutics SA. JDH: Cofounder of Visiopharm A/S. TE: Employee of Visiopharm A/S. ME: Egyptian
missions sector. WMG: Cofounder, shareholder and part‐time Chief ScientificOfficer of OncoAssure Limited, shareholder in Deciphex, and member of
scientific advisory board of Carrick Therapeutics. JMG: Employee and stockholder of Roche/Genentech. SG: Research funding from Regeneron
Pharmaceuticals, Boehringer Ingelheim, Bristol Myers Squibb, Celgene, Genentech, EMD Serono, Pfizer, and Takeda, unrelated to the current work;
named co‐inventor on an issued patent for multiplex immunohistochemistry to characterize tumors and treatment responses. The technology is filed
through Icahn School of Medicine at Mount Sinai (ISMMS) and is currently unlicensed. NH: Patent on a technology to measure immune infiltration in
cancer to predict treatment outcome (WO2012038068A2). MGH: Consultant for PaigeAI, VolastraTx, and advisor for PathPresenter. JH: Speaker's
honoraria or advisory board remunerations from Roche, Novartis, AstraZeneca, Eli Lilly, and MSD. Cofounder and shareholder of Stratipath AB. AIH:
Research fund received from Visiopharm A/S. KK: Employee and stockholder of Roche. AK: Honorarium from Roche, MSD, and Pfizer and is a member
of the advisory board of Pfizer. A‐VL: Institutional grants from AstraZeneca and personal grants from AstraZeneca (travel and honorarium from
advisory board), MSD (honorarium from advisory board), and Daiichi Sankyo (travel). XL: Eli Lilly Company, Advisor, Cancer Expert Now, Advisor,
Champions Oncology, Research fund. AM: Equity holder in Picture Health, Elucid Bioimaging, and Inspirata Inc., advisory board of Picture Health, Aiforia
Inc., and SimBioSys, Consultant for SimBioSys, sponsored research agreements with AstraZeneca, Boehringer‐Ingelheim, Eli‐Lilly, and Bristol Myers‐
Squibb, technology licensed to Picture Health and Elucid Bioimaging, involovement in three different R01 grants with Inspirata Inc. DKM: Consulting:
Astrazeneca, Lilly USA LLC, Hologic. Sponsored research: Merck, Agendia. SM: Scientific Committee Study member: Roche, data and safety monitoring
member of clinical trials: Sensorion, Biophytis, Servier, IQVIA, Yuhan, Kedrion. FuAAM: Research studentship funding from GSK. DAM: Speaker fees from
AstraZeneca, Eli Lilly, and Takeda, consultancy fees from AstraZeneca, Thermo Fisher, Takeda, Amgen, Janssen, MIM Software, Bristol‐Myers Squibb,
and Eli Lilly and has received educational support from Takeda and Amgen. FP‐L: Personal financial interests: AbbVie, Agendia, Amgen, Astellas,
AstraZeneca, Bayer, BMS, Daiichi‐Sankyo, Eisai, Exact Science, GSK, Illumina, Incyte, Janssen, Lilly, MERCK Lifa, Merck‐MSD, Myriad, Novartis, Pfizer,
Pierre‐Fabre, Roche, Sanofi, Seagen, Takeda, Veracyte, Servier. Institutional financial interests: AstraZeneca, Bayer, BMS, MSD, Myriad, Roche, Veracyte.
Congress invitations: AbbVie, Amgen, AstraZeneca, Bayer, BMS, Gilead, MSD, Novartis, Roche, Lilly, Pfizer. NMR: CoFounder, director and CSO of
4 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

Histofy Ltd, UK. JSR‐F: JSR‐F is an Associate Editor of The Journal of Pathology; he reports receiving personal/consultancy fees from Goldman Sachs, Bain
Capital, REPARE Therapeutics, Saga Diagnostics, and Paige.AI, membership in scientific advisory boards of VolitionRx, REPARE Therapeutics, and
Paige.AI, membership on the board of directors of Grupo Oncoclinicas, and ad hoc membership on the scientific advisory boards of Astrazeneca, Merck,
Daiichi Sankyo, Roche Tissue Diagnostics, and Personalis, outside the scope of this study. ES: Employee of BMS. AS: AS: Advisory board/speaker's
bureau: Aignostics, Astra Zeneca, Bayer, BMS, Eli Lilly, Illumina, Incyte, Janssen, MSD, Novartis, Pfizer, Roche, Seagen, Takeda, and Thermo Fisher, as
well as grants from Bayer, BMS, Chugai, and Incyte. FS: Expert advisory panel for AXDEV Group. TT: Employee of Tempus Labs. JT: Shareholder of
Ellogon.AI BV. TT: Speaker's fee received from Pfizer. JvdL: Member of advisory boards of Philips, the Netherlands, and ContextVision, Sweden, and
received research funding from Philips, the Netherlands, ContextVision, Sweden, and Sectra, Sweden, in the last 5 years. He is chief scientificofficer
(CSO) and shareholder of Aiosyn BV, the Netherlands. TW: Collaboration with the company TRIBUN Health on automatic grading of biopsies for head
and neck cancer and a patent on the prediction of homologous recombination deficiency (HRD) in breast cancer. YW: Employee of CellCarta. HYW:
Advisory faculty of AstraZeneca. YY: Speaker/consultant for Roche and Merck. PS: Consultant (uncompensated) to Roche‐Genentech. SL: Research
funding to her institution from Novartis, Bristol‐Meyers Squibb, Merck, Puma Biotechnology, Eli Lilly, Nektar Therapeutics, Astra Zeneca, Roche‐
Genentech, and Seattle Genetics. SLoi has acted as consultant (not compensated) to Seattle Genetics, Novartis, Bristol‐Meyers Squibb, Merck,
AstraZeneca, Eli Lilly, Pfizer, and Roche‐Genentech. SLoi has acted as consultant (paid to her institution) to Aduro Biotech, Novartis, GlaxoSmithKline,
Roche‐Genentech, Astra Zeneca, Silverback Therapeutics, G1 Therapeutics, PUMA Biotechnologies, Pfizer, Gilead Therapeutics, Seattle Genetics,
Daiichi‐Sankyo, Amunix, Tallac therapeutics, Eli Lilly, and Bristol‐Meyers Squibb. RS: Nonfinancial support from Merck and Bristol Myers Squibb (BMS),
research support from Merck, Puma Biotechnology and Roche, and personal fees from Roche, BMS, and Exact Sciences for advisory boards.
Introduction
The prognostic and predictive significance of the
tumor-immune interaction in breast cancer (BC) has
been investigated intensively in recent years [1,2],
and tumor-infiltrating lymphocytes (TILs) have
emerged as a robust biomarker with reasonable repro-
ducibility [3–5]. Within BC, triple-negative BC
(TNBC) (estrogen receptor, progesterone receptor,
and HER2-negative) and HER2-positive BC exhibit a
more pronounced tumor-associated immune cell infil-
trate. There is good evidence to suggest both a prog-
nostic and predictive potential for TILs in TNBC, even
in the absence of systemic chemotherapy [6]. In
TNBC, each 10% increment in TIL is associates with
a 17% relative increase in overall survival (OS) [7],
and TILs can predict chemotherapy response [8,9].
Therefore, routine evaluation of TILs during diagnos-
tic workup of TNBC patients was recommended in
the 2019 St Gallen International Breast Cancer
Consensus [10], and TIL assessment is now incorpo-
rated into several national guidelines as a biomarker
for TNBC and HER2-positive BC and is used
prognostically until predictiveness is validated in new
trials [11,12].
To move TIL evaluation from research and single-
center clinical use to routine cancer care, the clinical
evaluation must be accurate and reproducible. The
International Immuno-Oncology Biomarker Working
Group on Breast Cancer (also called the TILs-WG:
www.tilsinbreastcancer.org) has formulated a set of
guidelines for visual TIL assessment (VTA) on hema-
toxylin and eosin (H&E)-stained slides [13]. Although
this method of analyzing TILs is reproducible among
trained pathologists [14,15], there remains a need for
additional training, particularly because tumor hetero-
geneity affects reproducibility [16]. For this reason,
the US Food and Drug Administration (FDA) recently
provided an online publicly available continuing med-
ical education (CME) accredited TIL training course
for pathologists (https://ceportal.fda.gov/).
Recent developments in machine learning (ML) have
had a major impact on computational pathology [17],
including automated evaluation of TILs using ML and
image analysis –also referred to as computational TIL
assessment (CTA). CTA is a promising solution for
many of the issues of VTA and may lead to a standard-
ized and more reliable evaluation of TILs to complement
local TIL assesment when needed.
Using ML and digital image analysis to analyze
immune cell infiltration is not a new idea, having been
studied sporadically for the last decade or so, mainly
employing immunohistochemistry (IHC) [18–21].
Several novel approaches have demonstrated the prom-
ise of deep neural network-based algorithms for this task
on H&E stains [22–24]. However, important issues must
be taken into consideration during the development of
algorithms to evaluate TILs in BC, and new research,
development, and validation are required before ML
tools can be incorporated into the routine clinical man-
agement of BC.
In this review, we provide a perspective of the cur-
rent state of CTA and focus on how pitfalls with man-
ual assessment [14] also impact ML-based methods.
This is achieved by categorizing the inconsistent cases
reported in recent studies [22–24], and we extend the
analysis to include the unique challenges involved in
solving pitfalls with automated TIL evaluation. We
group our findings into four main areas: (1) general
pathology pitfalls, (2) ML and image analysis, (3) data
challenges, and (4) validation.
Background
In the TIL-WG VTA guidelines [13], TILs are defined as
mononuclear immune cells, lymphocytes, and plasma
cells. Intratumoral TILs (iTILs) that are in direct contact
with tumor cells are distinguished from stromal TILs
(sTILs) located in the stromal tissue between tumor cells
islands. The guidelines recommend focusing on sTILs
because their evaluation is more reproducible [7].
Pitfalls in ML assessment of TILs 5
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

sTILs are assessed as the ratio of the area occupied by
sTILs divided by the total tumor-associated stromal area,
with the final score reported as a percentage value. It is
imperative that areas of necrosis, ductal, and lobular
carcinoma in situ (DCIS/LCIS), and normal breast tissue
are excluded from the analysis.
The TIL-WG has reported on how computational
assessment of TILs could be designed, with the recom-
mendation that ‘computational TILs assessment (CTA)
algorithms need to account for the complexity involved
in TIL-scoring procedures, and to closely follow guide-
lines for visual assessment where appropriate’[25].
Several approaches to CTA can be considered, from
more granular approaches, closely mimicking the guide-
lines recommended by TIL-WG, to coarser strategies,
with methods also varying in their level of automation.
In this article, we focus predominantly on recently cre-
ated CTA algorithms that adhere to the guidelines.
Bai et al [23] produced an algorithm using the open-
source software QuPath [26], in which inclusion of
tumor regions and exclusion of noninvasive epithelium
(DCIS/LCIS and normal ducts and lobules) are manually
annotated by a specialist pathologist. Therefore, the
algorithm relies on an experienced pathologist since it
cannot identify the correct areas for analysis, nor can it
eliminate common artifacts [23]. This algorithm applies
color normalization before cells are segmented using a
traditional image analysis algorithm (background sub-
traction, thresholding, and watershed) to compensate for
H&E variability. Finally, a model trained on extracted
handcrafted cellular features classifies all cells as tumor
cells, TILs, fibroblasts, or others. The algorithm then
outputs five quantitative variables with prognostic sig-
nificance, including the TIL-WG definition: (1) total
area of TILs within the stroma (percentage); (2) number
of TILs in the annotated region; (3) amount of stromal
cells in the annotated region; (4) the total number of cells
in the annotated region; and (5) the proportional number
of TILs relative to the tumor [13].
Sun et al [24] presented a more comprehensive
approach, although still relying on manually annotated
regions by a pathologist. After identifying tumor regions
and excluding noninvasive regions, a tissue-level model
automatically detects and excludes necrosis to ensure that
necrotic cells are not misclassified as lymphocytes. Cells
are subsequently detected and classified as malignant epi-
thelial cells, TILs, or others using a cell-level model. The
classified cell algorithms are then used to identify the
tumor-, stroma-, and lymphocyte-dense regions using a
rule-based system, which produces a regional-based quan-
titative variable of the area coverage of sTILs.
Thagaard et al [22] reported a fully automatic system
using commercial software (Visiopharm A/S, Hørsholm,
Denmark), in which a tissue-level model identifies the
tissue types and then, with no manual interaction, auto-
matically identifies the invasive tumor, noninvasive
breast structures, and stromal and necrotic regions. A
cell-level model then identifies TILs and reports sTIL
density as a quantitative variable. Other studies have
proposed alternative metrics [21,27,28] or used stains
other than H&E [29–31]. We have found that these
methods show inconsistency with the TIL-WG VTA
guidelines (see [25]).
The common findings from these studies are that CTA
has good to excellent agreement with VTA and, more
importantly, is independently associated with clinical
outcome, confirming that patients with TNBC and a high
CTA score have improved survival [22,24]. In addition,
the studies indicate that current CTA is not a panacea for
the limitations of VTA, and more research is needed to
address handling pitfalls, along with further develop-
ment and clinical validation of CTA.
Common pitfalls between visual and computational
assessments
On behalf of the TIL-WG, Kos et al [14] identified and
reported the most common pitfalls of evaluating TILs by
eye. Some are also relevant when developing ML
approaches and will be discussed here, as will those
unique to CTA.
Including wrong areas or cells
The most frequent cause of inconsistent CTA results
compared to manual scoring is the inclusion of incorrect
areas for evaluation. These tissue-level pitfalls include
(1) TILs around noninvasive structures (DCIS/LCIS,
benign lesions, and normal ducts and lobules)
(Figure 1)[22,24]; (2) lymphocytes associated with
other structures (such as lymphovascular invasion and
vessels in general; Figure 1)[24]; (3) necrotic areas [24];
and (4) tertiary lymphoid structures (TLSs), possibly as
an aggregate because H&E staining does not allow differ-
entiation between B- and T-cells [23]. In addition, training
algorithms are commonly developed on ductal tumors,
making potential pitfalls for lobular histology and less
common histologic subtypes such as mucinous, metaplas-
tic, apocrine, and papillary cancers [22,24]. Both Sun et al
and Bai et al suggested excluding these confounding
regions manually, which is therefore subject to the same
pitfalls as full VTA and reduces time efficiency due to
pathologist involvement [23,24]. In Thagaard et al this step
was performed automatically, with issues for complex pat-
tern equivocal DCIS but not benign regions or uniform
DCIS. Overall, manual and automatic approaches have
the same pitfalls regarding regions of equivocal DCIS.
The extent to which this impacts the accuracy of computa-
tional tools for TILs has yet to be fully resolved [22].
Cell-level problems where incorrect cell detections
are included are less prevalent in CTA. Bai et al reported
substantial segmentation failure in 1–2%, i.e. where the
segmentation model is the major cause of discordant
cases. The main cause was an inability to distinguish
iTILs from sTILs, causing tumors with a high proportion
of iTILs to be excluded from the study. Apoptotic bod-
ies, neutrophils, and low-grade or neuroendocrine
tumors can also lead to false-positive TIL detection [23].
6 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

The aforementioned performance pitfalls can be
addressed using approaches described in the subsequent
section ‘Image analysis challenges when adhering to a
clinical guideline’and by using variable data for model
training described in ‘Training data challenges to create
robust and generalizable algorithms’.
Technical factors that impact ML algorithms
Slide-related issues are a common challenge for VTA [14]
and also impact CTA. Variables in the preanalytic
workflow include cautery artifacts, as well as tissue dyes
used to mark resection margins during macroscopic exam-
ination. Artifacts of histological preparations include those
derived from tissue processing, microtomy, staining, and
mounting (zonal fixation, blade lines, tissue disruption,
microchatters, air bubbles, floaters). Out-of-focus areas,
pen markings, tissue folds, blurring, air bubbles, thick
sections, and crush artifacts can each confuse tissue- or
cell-level models and consequently lead to inaccurate
quantification. For example, poor sectioning can cause
false-negative TILs, thereby producing an underestimation
of the true TIL density [22].
Scanning variability among different manufacturers is
also a problem when comparing cohorts of multi-
institutional studies because of the lack of standardized
acquisition parameters. The extent of this issue in CTA has
yet to be properly investigated [22], but for applications
such as detection of prostate cancer, variation influences
the uniform interpretation of CTA. Similarly, inter- and
intrasite variation in slide preparation and/or staining may
contribute to differences in CTA between cohorts [23,24],
similar to other applications [32].
There are two main approaches to combating these
problems. First, they can be handled manually or by
employing a separate model, e.g. excluding out-of-focus
cells [33]orfields [34] in a preanalysis phase. Second,
more variability in scanning and staining quality metrics
can be incorporated into the dataset used to develop
CTA, and we cover key aspects of these issues in what
Figure 1. Lymphocyte-dense regions associated with other structures should be excluded as the inflammation is not necessarily an immune
response to the tumor. (A) TLS. (B) Lymphocytes surrounding vessels. These areas are reported [24] as possible false-positive areas in CTA at
much higher levels than VTA. Images by Elisabeth Specht Stovgaard from Herlev cohort used in [22].
Pitfalls in ML assessment of TILs 7
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

follows (‘Training data challenges to create robust and
generalizable algorithms’).
Heterogeneity in sTIL distribution and
tumor-compartment definitions
One pitfall that causes the highest manual interobserver
variation is the presence of increased sTILs at the lead-
ing edge of a tumor compared to the central tumor
area [14] (Figure 2A). This aspect of CTA compared to
VTA was highlighted in recent studies [22,24]. The
increased density of sTILs at a tumor’s leading edge
can cause a lower CTA score, because the immune-
deserted stromal region in the central tumor region will
contribute to a larger stromal area quantification than
would be estimated by manual assessment. In contrast,
if stroma is scarce in the central tumor, the high-density
margin will contribute most to the overall score,
resulting in a higher CTA score.
In general, the identification of tumor-associated stro-
mal regions in which sTILs should be scored is not
strictly defined in manual guidelines. Sometimes there
are larger stromal areas within the tumor core, but the
allowable distance between stromal TILs and tumor
nests for quantitation remains unclear (Figure 2B).
Similarly, there is no quantitative definition of outlier
TIL density hotspots that should be excluded. This can
lead to discrepancies between VTA and CTA [22],
depending on the CTA integration details method and
the validation approach employed (discussed further in
the sections ‘Image analysis challenges when adhering
to a clinical guideline’and ‘Validation challenges when
comparing CTA with VTA’).
Moving beyond human capabilities
Some of the aforementioned discrepancies may be elim-
inated by algorithm improvements, such as more accu-
rate delineation of stroma [24]. However, other pitfalls
may arise when tissue-level outlining becomes too
precise [22]. Specifically, CTA detects very small areas
of stroma within tumor nests, which a pathologist might
not consider due to their size, but no rules exist to define
how small a stromal area can be to be included. This
problem can lead to higher or lower measurements of
TILs than a manual score if these areas include many
TILs (larger TIL count) or do not include TILs (larger
stromal area). The highly accurate quantification
allowed by CTA will therefore lead to discrepancies
with VTA pathologic evaluation; the gold standard will
be the method that provides the highest clinical benefit,
measured by its predictive or prognostic accuracy [22].
The choice between standard VTA and CTA-derived
guidelines will be settled by discrepancy aspects, which
is discussed subsequently.
Image analysis challenges when adhering to a
clinical guideline
Many of the pitfalls mentioned can be attributed to the
image analysis approach that is used to implement the
rules of the VTA guideline. However, the gold reference
for scoring most existing histology-based biomarkers is
currently the pathologists’assessment, for instance,
HER2 [34] and the VTA guideline [13]. Hence, the
computational pathology community always needs to
Figure 2. Examples of discrepant cases from Herlev cohort used in [22]; purple areas: tumor nests, heatmap areas: sTIL regions. (A) A case of
high sTIL density at tumor margin compared to central area. As the stroma is scarce inside the tumor, sTIL density is reported to be very high in
CTA as mostly the margin contributes to the score. (B) The tumor grows irregularly with small tumor nests between larger invasive tumor
areas. In these cases, the CTA includes more stroma than VTA, resulting in a lower sTIL density score (larger denominator) than the manual
score.
8 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

answer the same question first: What strategy do we
want to use to translate the rules of manual guidelines
into something a computer can execute? There are many
valid answers to this question and we review the pros
and cons of CTA in the following sections.
Different approaches to the computer vision
problems relevant for building interpretable TIL
algorithms
Here, we focus on three main categories of computa-
tional approaches for quantifying tissues and cells:
(1) patch classification, (2) object detection, and
(3) image segmentation. Although several other methods
exist, we focus on the most commonly used approaches
(see [35] for a more extensive review).
First, the simplest approach is classification, some-
times referred to as patch-based approaches in digital
pathology, especially for methods employing deep
learning models. A supervised learning algorithm incor-
porates an image patch/field of view (FOV) for classifi-
cation into one discrete label/class from a predefined set
of labels. Second, object detection extends classification
by producing one class per object of an image along with
its spatial location. The location of objects of interest is
marked with a box around the detected object, or the
object is marked at its center. Finally, segmentation goes
beyond detection by assigning a label/class to every
pixel in the image to create a semantic map of the types
of objects and their location. In contrast to object detec-
tion, segmentation can outline the border of the objects
at very high resolution and accuracy. The general con-
sideration when selecting a category for a computer
vision problem is to determine the level of precision/
resolution needed (i.e. how coarse the output can be to
still adhere to the guideline). There are also differences
in terms of training data and validation requirements that
we will cover in later sections.
According to the TIL-WG guideline [24], a CTA
algorithm should be able to (1) detect and compartmen-
talize tissue into tumor structures, tumor-associated
stroma, and stroma outside tumor borders and (2) quan-
tify TILs in the compartment. Different considerations
apply to these two topics.
Considerations for tissue-level models
Object detection is not suited for subdividing complex
tissue structures into distinct areas (e.g. highly infiltrat-
ing tumor nests) and is therefore often excluded for the
recognition and compartmentalization of tissue. Most
CTA algorithms use a classification approach [21,27]
or full segmentation [20,36]. The main difference is the
granularity of the annotated maps produced, with seg-
mentation being finer than classification, although reso-
lution varies depending on the overlap and tile size of the
input image patches and the size of the sliding window.
When using a direct classification of image patches as
tumor, stromal, or lymphocyte regions, an individual
patch may contain different tissue components, making
classification not only difficult to train because of the
inherent noise in the annotations but also imprecise for
prediction due to the presence of multiple classes in a
single image for only one output class. Moreover, a
patch-based approach may not provide detailed, quanti-
tative information on TIL density; for instance, an accu-
rate patch-based lymphocyte classifier would produce the
same output whether only one or many lymphocytes are
within an input image. Bai et al [23] used manual
outlining by a pathologist and did not discriminate
between tumor and stromal areas, which was problem-
atic for high-iTIL cases, as mentioned previously. Sun
et al [24] also used manual outlining in combination
with a patch-based model to identify and exclude
necrosis. They then used the cell-level output (see the
next section for details) and empirically defined a tumor
area as a patch containing more than two tumor cells.
This sliding window approach produced relatively
coarse boundaries compared to a full segmentation
model [20,22].
Providing models that segment the tissue allows for
the construction of more detailed and quantitative infor-
mation at the cellular level. Even though segmentation
seems the obvious choice to carry out the tissue-level
task of detecting the stromal areas necessary for CTA,
the approach also has disadvantages. First are potential
segmentation artifacts from a sliding window analysis,
which is preferred due to the gigapixel size of whole-
slide images (WSIs). This can also lead to tiling-induced
issues that result in incorrect labeling/misleading cate-
gorization (e.g. one glandular structure is divided in two,
and these are analyzed independently, with one part
being segmented as invasive tumor and the other part
as DCIS). In the naïve setup, the ML model only takes
into account one part at a time in what is called the
receptive field (i.e. the tissue structure that the model
sees at each prediction). Such inconsistencies along the
edges of each FOV need to be handled, and in this
postprocessing strategies can be helpful. If two segments
of DCIS and invasive tumor regions touch as a single
object, the size and shape of the DCIS segment can be
considered in a logical postprocessing step to determine
whether both should be segmented as DCIS or invasive
tumor [22]. The important point is that these events are
handled consistently, and with relevance to the clinical
guideline. For example, one should rather exclude DCIS
because there is often a high density of stromal TILs
around these preinvasive lesions, and including false-
positive regions around DCIS structures would heavily
influence the overall TIL score. Systematic inclusion of
clinical guidelines in the digital framework is needed at
either the data preprocessing or postprocessing stage.
Considerations for cell-level models
The objective of TIL quantitation is to output the per-
centage of TILs in a given tissue region. This step is
usually performed at the same or higher magnification
than is used for tissue-level analysis to include sufficient
cell-level image features for accurate model predictions.
Pitfalls in ML assessment of TILs 9
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

The main goal is to distinguish mononuclear immune
cells from other cells. Previous studies performed this
task by classification, detection, and segmentation.
Janowczyk and Madabushi [37] used a classification
model with a small sliding window to obtain the most
likely location of each lymphocyte. A potential draw-
back of this method is computational inefficiency, as its
high precision requires highly overlapping predictions.
More recently, several studies [22,38,39] employed seg-
mentation models to directly predict the center of all
TILs in a FOV, avoiding the latter inefficiency issue.
Others [24,40] used a combination of both object detec-
tion and segmentation [41] to obtain the location and
outline of TILs simultaneously. There are minor differ-
ences between the methods; the main challenges for cell-
level models relate to the requirements of the training
data required for their development, which we discuss in
the following section. Future work should also investi-
gate improving the accuracy of cell classification to guide
the model with information about the location of cells
derived from compartment classification/segmentation.
The idea is that the probability that cells will belong to a
particular category (e.g. lymphocytes, fibroblasts)
depends on their location in tissue compartments (like
epithelium or stroma).
Another consideration here is the definition of the
final sTIL score as a quantitative output variable, and
recent methods have used different definitions. The VTA
guideline uses an area coverage approach, which is the
most accessible for humans to estimate. However, this
introduces a slight size bias toward larger TIL nuclei.
Does this mean that a CTA should do the same? We
argue that as long as the CTA quantifies the degree of
immune cell infiltration and is interpretable by patholo-
gists (either by heatmaps or a score), then it is a valid
score, and validation methods will then identify the most
appropriate scoring system. That recent papers [22–24]
found seven output variables associated with survival is
evidence for this. Interestingly, although the VTA guide-
line explicitly states that sTILs should not be scored as a
fraction of TILs compared to other cell populations, two
variables of this assessment type consistently provide
better results [23]. Thus, there might be other ways of
creating a CTA, but it could also just be a derivative of
the model design proposed in that paper.
Training data challenges to create robust and
generalizable algorithms
The described models are exclusively built using deep
learning, a powerful form of ML that, given sufficient
training examples, learns to unravel and identify com-
plex patterns. We will not review all aspects of this field
but instead refer the reader to other excellent review
articles [35,42]. However, since the most promising
CTA algorithms use deep learning, we will cover one
of the main challenges of creating such algorithms:
obtaining the training data required.
Data variation considerations
The general rule for creating a development/training
dataset (i.e. the data used to develop the algorithm) is
to include as much interclinical variation as the algo-
rithm can be expected to encounter. Therefore, the
requirements depend on the scope of the CTA algorithm,
meaning the level of generalization required. For
instance, single-center research studies are deployed in
only one laboratory. In a multicenter study, different
laboratories participating in the training relate to internal
validation, or a laboratory outside of model development
is related to external validation. The answer to these
questions indicates what boundaries of variation the
CTA algorithm is expected to handle. The main sources
of variation originate from the significant challenges in
standardization within pathology. As such, before begin-
ning image analysis, quality control of tissue, histology
slide, stain, and WSI should be confirmed to ensure that
a standard is met that will allow the collection of reliable
data. Variability across pathology laboratories in
preanalytical (e.g. fixation, sectioning) and analytical
(e.g. staining protocol, scanner model) variables causes
distributional shifts in the image data. Studies have
investigated the impact of such variables, and methods
to normalize and/or decrease variability from scanners
[43–45] and staining [32,46] have been developed.
Another important factor when curating a dataset is
the impact of histological subtype variability (invasive
ductal, invasive lobular, mucinous) on the underlying
data distribution. Even the most powerful computational
models, such as deep learning, may not generalize out-
side the subtype seen during training [47,48]–one
should not expect to successfully implement a model
on lobular carcinoma if the data used for model devel-
opment include only ductal carcinomas. This aspect sets
some requirements on how to source and sample the
patient cohort as part of relevant inclusion and exclusion
criteria in the study design and should yield a balanced
and realistic dataset. It is important to remember that for
any digital model to work in a generalizable manner,
interclass (between-group) variation must be higher than
the intraclass (within-group) variance.
Generally, the solution to these issues is straightfor-
ward. Simply including relevant and sufficient variation
in the development dataset aids in making the algorithm
robust and generalizable. But even with increased sam-
ple numbers, the training set will only partially represent
the full data distribution, and the trained algorithm will
therefore be confronted with some previously unseen
situations during application. Methods to identify, mon-
itor, and flag additional novel classes [47], dataset shifts
[48,49], and normalization schemes [32,43,46] should
help to reduce this problem.
Data labeling considerations
Acquiring an adequate number of manual labels is a
critical step in computational pathology, given the time
and effort required from pathologists and others with the
specific expertise required. Several approaches have
10 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

been proposed to address the need for manual labels in
large-scale datasets. The requirements for time, exper-
tise, and methods depend on the model type being
trained. The magnitude of investment correlates with
the precision of the output required. In general, classifi-
cation labels are the simplest to obtain (only one value is
needed per image), then object detection labels (one
click and one label per object), and then segmentation
labels (many clicks and one label per object). For the
new approaches being proposed, the major objective is
to limit pathologist involvement to avoid the high cost,
the time constraints of clinical practice, and the repetitive
nature of annotating multiple examples.
The most straightforward strategy is manual annota-
tions by a large number of experts. This approach gen-
erates high-quality labels because ambiguous labels are
identified and corrected, but it remains expensive and
suffers from interlabeler variability and the subjective-
ness inherent in histopathology. One solution is to ask
multiple annotators to annotate the same data and pro-
duce a consensus label or model label variability [50]. A
crowdsourcing framework for both tissue-level segmen-
tation and cell-level classification, object detection, and
segmentation was proposed to reduce pathologist effort
and to model the interlabel variability of multiple
labelers [40,51]. Multiple nonpathologists (up to six)
were required to match the performance of a senior
pathologist. However, the benefit is restricted to anno-
tating predominant and visually distinctive patterns,
implying that pathologist involvement, and possibly
full-scale labeling effort, will be needed to supplement
uncommon and difficult classes that require expertise.
Of note, training and test sets must include borderline
cases that are encountered in real life but might be hard
to annotate. Otherwise, when trained and tested exclu-
sively on ‘clean’data, the algorithm may have difficul-
ties with data for which the decision is harder to
establish.
One of the most important aspects of developing a
labeled dataset for CTA is the consistency of labels and
annotations, i.e. minimization of ambiguous samples in
the dataset. This consistency is difficult to adhere to
when relying on manual labels. Compared to other
fields, such as radiology, histopathology is unique in
terms of creating a ground-truth definition. For many
applications, we rely on experts for ground truth, but we
can also use the antibody–antigen specificity of
immunohistochemical stains. Recently, multiple label-
ing schemes were proposed to obtain tissue- and
cell-level labels [22]. The idea is to use IHC to guide
semiautomatic labels that can be transferred to primary
H&E slides, and models can then be trained and
deployed on H&E only. The obvious pitfall is the need
to prepare new serial sections, which means using more
tissue. Also of relevance for TILs, cellular information
might be compromised between consecutive sections.
Alternatively, the H&E section can be restained if the
expertise is available, thereby ensuring that IHC-stained
lymphocytes can be found in the previously H&E-
stained slide. Even though this approach requires
additional developmental effort, the quality and consis-
tency of the labels were reported to be higher than those
of manual labels, and only one pathologist was needed to
review the labels, decreasing the time and effort required
for model development [22].
Because WSIs are gigapixel files, it is intractable to
manually label entire WSIs. Therefore, one needs to
sample training regions, where it is important to use
the same principles of including data (label) variation,
e.g. regions with low-, medium-, and high-density TILs
should be included, also with varying proximity to inva-
sive cells. One solution is to build weakly supervised
image segmentation models that do not require detailed
cell-level labels.
Although many schemes can be employed to optimize
the time and necessity for pathologist involvement, such
procedures have their own pitfalls, as discussed earlier.
We always advise developing an annotation protocol
and labeling strategy in collaboration with a pathologist
and treat it as an iterative process to identify errors and
inconsistencies that will enhance the quality and scale of
the training labels [52].
Data access and sharing considerations
It is obvious that access to raw data is a prerequisite for
developing CTA algorithms. However, there are substan-
tial challenges in collecting and/or accessing appropriate
sets of data. Not all laboratories systematically scan all
slides on modern scanners into an image management
system (IMS) or picture archiving and communication
system (PACS). Even fewer departments have digitized
their archived slides or, indeed, have enough computer
storage for such archiving. Scanning large retrospective
datasets is, on the one hand, time-consuming since most
scanners need to be manually checked for quality,
although on the other hand this may simplify future
research.
A key aspect of developing successful CTA algo-
rithms is collaboration between partners, among aca-
demic centers or academia and industry. The
development of such algorithms on WSIs will be
reviewed by a pathologist to communicate with the data
scientists to improve algorithm adjustment and ML
training. Another important aspect is analysis of the
pathologist’s notes. These notes and the complete
clinical-morphological data include information regard-
ing stage, molecular profile, previous biopsies, and post-
treatment changes. Such analysis might be performed
using a natural language processing pipeline to extract
data for further standardized reporting. Getting the legal
terms and conditions into place to share data can be a
lengthy process. Sharing is recommended because it
substantially eases this process for patient information
protection regulations and the included requirement on
information technology infrastructure and security.
Another important consideration is the size of the
datasets and how local or cloud platforms can be used
to store, access, and share data.
Pitfalls in ML assessment of TILs 11
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

There are successful studies sharing high-quality his-
tology datasets publicly under Creative Commons
(CC) licenses [53,54], either fully public [51,55]or
restricted for noncommercial use [56]. The latter can
hinder academia–industry collaborations. The most
commonly used platforms for public sharing of datasets
are the Grand Challenges website [57] and The Cancer
Genome Atlas (TCGA) [58]. Historically, there has been
a shortage of publicly available datasets for developing
CTA systems, with TCGA a notable exception and
providing the foundation for many CTA studies
[21,22,24,40]. Nonetheless, care must be taken to
avoid bias and batch effect implications from public
datasets, which were not necessarily created for TIL
evaluation [59]. There are recent joint efforts from the
FDA and TIL-WG to create datasets for algorithm
validation [50,60], to fill the critical need for the
availability of development datasets. Collecting a large
number of WSIs is time-consuming and is subject to
approval by institutional review boards and a data
protection officer to comply with privacy and patient
laws. Conversely, curating remains a barrier to the
scaling of CTA algorithms.
Validation challenges when comparing CTA
with VTA
Quantitative metrics on the performance of the different
parts of a CTA algorithm need to be evaluated during
development and, especially, during validation of the
image analysis model [61]. As previously reviewed by
the TIL-WG [24], there are different levels of perfor-
mance measurement. Briefly, analytical validation
(AV) refers to low-level metrics such as accuracy and
reproducibility; clinical validation (CV) describes the
discrimination of patients into clinical subgroups; and
clinical utility measures the overall benefit in a clinical
setting. In the following subsections, we discuss poten-
tial pitfalls in model validation.
Subcomponents of modular systems need different
evaluation metrics
It is clear that to adhere to the TILS-WG guideline, an
accurate CTA algorithm must consist of multiple models
aimed at solving different parts of the guideline. Hence,
AV applies to the subcomponents as well as to the entire
system. As the subcomponents can be different model
approaches, the AV metric needs to capture aspects of
each approach while providing information in situations
where failure of a subcomponent will cause failure of
CTA. Metrics such as accuracy, precision, recall,
F-scores, and Matthew’s correlation coefficient are some
of the other measures used to evaluate model performance.
If a subcomponent is a segmentation model (e.g. the
tumor, necrosis, and noninvasive tissue-level model),
standardized metrics such as the F1 score can be used
to evaluate AV. The F1 score can be interpreted as the
weighted average of the precision and recall/sensitivity.
However, it is important to consider that the F1 score on
a FOV with no true-positive segments of any given class
will be evaluated as zero for that class, implying that
potential false positives will not be captured as false
positives, invalidating the overall F1 score. Another
challenge for subcomponent AV is the impact of the
exact test score of the model. A benchmark for the exact
model selection does not always exist; hence, it is diffi-
cult to know if an exact score is sufficient or if a better
(or worse) model would impact the AV and/or CV of
the CTA.
Dudgeon et al [50] proposed both a metric and a
dataset that might qualify as a FDA Medical Device
Development Tool [60]. The metric is a multireader,
multicase version of the mean squared error. Similar
metrics such as Spearman rank-based correlation are
often used for the algorithm-to-pathologist comparison
[20,22]. One of the pitfalls of such count-based metrics
is that they do not capture whether the pathologist and
algorithm are counting the same or different TILs
because they compare only the sum of TILs. However,
the metrics are easy to use and interpret, and they capture
the most clinically relevant aspect of the algorithm –the
extent of TILs in a defined region.
Considerations regarding clinical validation and
utility
For the AV of the full algorithm, the same metrics can be
used for the algorithm-to-pathologist comparison.
However, as recently commented [62], the best method
to evaluate digitally assessed biomarkers, such as CTA
for both AV and CV, remains an open question. This
point to the paradox of selecting the ground truth for
digital pathology in TILs as either concordance between
the pathologist and computational score or patient out-
come, or a combination of both. This also raises the
question of clinical cut-off value for sTILs, since there
are no formal recommendations at this time. The lack of
manual VTA-based cut-off for patient stratification into
clinically meaningful subgroups makes the process of
CV more challenging for CTA because any cut-off com-
parison between VTA and CTA might be arbitrary.
Current CTA studies [22–24] use other cut-off points
than those used for VTA [3,62–64] to identify two
patient groups (TILs-high versus TILs-low) and find
different levels of agreement between manual and auto-
mated methods at different cut-offs. Sun et al [24] found
moderate to substantial agreement depending on the
exact cut-offs, but only moderate agreement at a 10%
cut-off. In contrast, a different cohort [22] showed sub-
stantial agreement at 10% cut-off. Interestingly, the for-
mer findings might imply different TIL cut-off values are
important, depending on the cohort and patient ethnici-
ties, although no significant difference in TIL distribu-
tion was found between Asians and Caucasians [24].
This highlights the general difficulties of finding a cut-
off for biomarkers, which still involves a high degree of
uncertainty [62]. In contrast, both studies found that
12 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

CTA score as a continuous variable was associated with
disease-free survival (DFS) and OS. Hence, TILs could
be better integrated into prognostic modeling containing
existing clinical variables such as age, lymph node
status, tumor size, tumor grade, and tumor type,
removing the need to determine a cut-off even for dif-
ferent ethnicities. The optimal method of TIL
assessment –threshold versus continuous may be
different for VTA versus CTA and remains an area of
active research, e.g. against alternative endpoints such as
BC progression [65].
Discussion
Current state-of-art CTA algorithms suggest that sTILs
can be assessed computationally and represent a crucial
prognostic and predictive factor for TNBC [7]. This
review highlights different methodological approaches to
designing algorithms. Beyond methodological design,
many of the same pitfalls exist for VTA [14]. Whether
these influence the clinical validation of CTA is to be
determined, given that it depends on the future approaches
taken to validate these algorithms. The TIL-WG is cur-
rently organizing a grand challenge using phase 3 clinical
trial data, which is a crucial step in validating any CTA
algorithm [62]. This may answer many of the questions
related to the clinical importance of CTA precision that are
currently difficult to evaluate. However, similar collabora-
tive community-driven initiatives are needed to create
robust and generalizable CTA algorithms. Many techno-
logical and procedural standardizations and harmoniza-
tions are necessary to counteract model-decay and
interinstitutional differences in workflow, especially in
difficult tissues (e.g. small, deformed morphology or poor
tissue integrity). Currently, there is no public framework or
infrastructure to work collaboratively on different labeling
strategies ensuring that CTA algorithms can identify and
handle all histological components, including DCIS, fibro-
sis, hyalinization, and a larger number of granulocytes.
There is currently no easy and practical way of building
combined versioned datasets of standardized WSI and
label formats, largely due to institutional data-sharing
restrictions and privacy requirements.
Another important unresolved aspect is the human–
algorithm interaction, i.e. when and how the algorithm
should be introduced into the workflow. Should the
pathologist be required to open a case and manually
annotate or edit regions, send the case for analysis, and
wait for the result? Or should the algorithm be auto-
mated so that a case is analyzed based on slide metadata
readily available after scanning, meaning that the case
will have already been analyzed when the pathologist
opens it for the first time? We deem the former unreal-
istic due to time and workload constraints. Different
implementations will need to be optimized to augment
and not disrupt the current workflow. Similarly,
uncertainties remain on the best way to present the
quantitative results of CTA, e.g. a precise count of
TILs per square millimeter or a relative area. A dichot-
omous score of both computational and manual mea-
surement may predict outcomes better than either
variable alone [24]. This might affect whether the
CTA should provide the primary score or work as a
secondary reader on difficult cases.
It is clear that CTA is a powerful tool, but it is
beneficial only when in the hands of expert pathologists.
Work is in progress on many of these challenges as we
look to an exciting future. Aware of the responsibility of
the pathologist’s decision-making, we hold as our ulti-
mate goal the development of robust tools for patholo-
gists that assist with personalized precision care in a
standardized and time-efficient manner. We hope that
by highlighting the specific pitfalls in using ML for sTIL
assessment during both the model development and the
clinical translation stages, future developments and col-
laborations will be positioned/forged to find the solu-
tions needed to ensure reliable computational reporting
of sTILs, with the end goal of using this tool in the
routine clinical management of BC.
Acknowledgements
The authors would like to thank Jeannette Parrodi, PA
assistant to Professor Sherene Loi, for het extensive help
and administrative support for the International Immuno-
Oncology Biomarker Working Group (TIL working
group). Without her, this working group would not even
exist. Furthermore, the authors make the following
acknowledgments regarding support and funding. GB:
Funded by Gilead Breast Cancer Research Grant 2023.
SV: Supported by Interne Fondsen KU Leuven/Internal
Funds KU Leuven. BA: supported by the Swedish
Society for Medical Research (Svenska Sällskapet för
Medicinsk Forskning) postdoctoral grant, Swedish
Breast Cancer Association (Bröstcancerförbundet)
Research grant 2021. GC: Peer Reviewed Cancer
Research Program (Award W81XWH-21-1-0160) from
the US Department of Defense and the Mayo Clinic
Breast Cancer SPORE grant P50 CA116201 from the
National Institutes of Health (NIH). CF-M: Funded by
the Horizon 2020 European Union Research and
Innovation Programme under the Marie Sklodowska
Curie Grant agreement No. 860627 (CLARIFY
Project). SBF: NHMRC GNT1193630. WMG: Support
by the Higher Education Authority, Department of
Further and Higher Education, Research, Innovation
and Science, and the Shared Island Fund [AICRIstart:
A Foundation Stone for the All-Island Cancer Research
Institute (AICRI): Building Critical Mass in Precision
Cancer Medicine, https://www.aicri.org/aicristart]:
Irish Cancer Society (Collaborative Cancer Research
Centre BREAST-PREDICT; CCRC13GAL; https://
www.breastpredict.com), the Science Foundation
Ireland Investigator Programme (OPTi-PREDICT;
15/IA/3104), the Science Foundation Ireland Strategic
Partnership Programme (Precision Oncology Ireland;
Pitfalls in ML assessment of TILs 13
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

18/SPP/3522; https://www.precisiononcology.ie). SG:
Partially supported by NIH grants CA224319,
DK124165, CA263705, and CA196521. AG:
Supported by Breast Cancer Now (and their legacy char-
ity Breakthrough Breast Cancer) and Cancer Research
UK (CRUK/07/012, KCL-BCN-Q3). TRK: Japan
Society for the Promotion of Science (JSPS)
KAKENHI (21K06909). UK: Funded by Horizon 2020
European Union Research and Innovation Programme
under the Marie Sklodowska Curie Grant agreement
860627 (CLARIFY Project). JKL: This work is in part
supported by NIH R37 CA225655 to JKL. AM: Research
reported in this publication was supported by the National
Cancer Institute under award numbers R01CA268287A1,
U01CA269181, R01CA26820701A1, R01CA249992-
01A1, R01CA202752-01A1, R01CA208236-01A1,
R01CA216579-01A1, R01CA220581-01A1, R01CA2
57612-01A1, 1U01CA239055-01, 1U01CA248226-01,
and 1U54CA254566-01, National Heart, Lung and
Blood Institute 1R01HL15127701A1, R01HL15807101
A1, National Institute of Biomedical Imaging and
Bioengineering 1R43EB028736-01, VA Merit Review
Award IBX004121A from the US Department of
Veterans Affairs Biomedical Laboratory Research and
Development Service the Office of the Assistant
Secretary of Defense for Health Affairs, through the
Breast Cancer Research Program (W81XWH-19-1-
0668), the Prostate Cancer Research Program (W81
XWH-20-1-0851), the Lung Cancer Research Program
(W81XWH-18-1-0440, W81XWH-20-1-0595), the Peer
Reviewed Cancer Research Program (W81XWH-18-1-
0404, W81XWH-21-1-0345, W81XWH-21-1-0160), the
Kidney Precision Medicine Project (KPMP) Glue Grant,
and sponsored research agreements from Bristol
Myers-Squibb, Boehringer-Ingelheim, Eli-Lilly, and
Astrazeneca. SKM: Kay Pogue-Geile, Director of
Molecular Profiling at NSABP for her constant support
and encouragement, Roberto Salgado, for initiating me
into the wonderful subject of Immuno-Oncology and
its possibilities. FuAAM: Funding from EPSRC
EP/W02909X/1 and PathLAKE consortium. FP-L:
Research grants from Fondation ARC, La Ligue contre le
Cancer. RDP: The Melbourne Research Scholarship and a
scholarship from the Peter MacCallum Cancer Centre.
JSR-F: Funded in part by the Breast Cancer Research
Foundation, by a Susan G. Komen Leadership grant,
and by the NIH/NCI grant P50 CA247749 01. JS:
NIH/NCI grants UH3CA225021 and U24CA215109.
ST: Supported by Interne Fondsen KU Leuven/Internal
Funds KU Leuven. JT: Supported by institutional grants of
the Dutch Cancer Society and the Dutch Ministry of
Health, Welfare and Sport. EAT: Breast Cancer Research
Foundation grant 22-161. GEV: Supported by Breast
Cancer Now (and their legacy charity Breakthrough
Breast Cancer) and Cancer Research UK (CRUK/07/01
2, KCL-BCN-Q3). TW: Support by the French govern-
ment under management of Agence Nationale de la
Recherche as part of the Investissements d’avenir’pro-
gram, reference ANR-19-P3IA-0001 (PRAIRIE 3IA
Institute), and by Q-Life (ANR-17-CONV-0005). HYW:
Funded in part by the NIH/NCI grant P50 CA247749 01.
YY: Funding from Cancer Research UK Career
Establishment Award (CRUK C45982/A21808). PS:
Funding support from the National Health and Medical
Research Council, Australia. SL: Supported by the
National Breast Cancer Foundation of Australia (NBCF)
(APP ID: EC-17-001), the Breast Cancer
Research Foundation, New York [BCRF (APP ID:
BCRF-21-102)], and a National Health and Medical
Council of Australia (NHMRC) Investigator Grant (APP
ID: 1162318). RS: Supported by the Breast Cancer
Research Foundation (BCRF, grant 17-194).
Author contributions statement
JT, GB and ES conceptualized, developed methodology
and wrote the original draft. JT and ES were responsible
for visualization. All authors were involved in reviewing
and editing the original draft. SH, AD, TE, JD, EB and
RS supervised. ZK, GA, NB, FC, EH, MK, RM, FP,
JMR and ES were involved in writing, reviewing and
editing the original draft. All authors have read and
agreed to publish the final version of the manuscript.
References
1. Bates GJ, Fox SB, Han C, et al. Quantification of regulatory T cells
enables the identification of high-risk breast cancer patients and those
at risk of late relapse. J Clin Oncol 2006; 24: 5373–5380.
2. Wang M, Zhang C, Song Y, et al. Mechanism of immune evasion in
breast cancer. Onco Targets Ther 2017; 10: 1561–1573.
3. Savas P, Salgado R, Denkert C, et al. Clinical relevance of host
immunity in breast cancer: from TILs to the clinic. Nat Rev Clin
Oncol 2016; 13: 228–241.
4. Hammerl D, Smid M, Timmermans AM, et al. Breast cancer geno-
mics and immuno-oncological markers to guide immune therapies.
Semin Cancer Biol 2018; 52: 178–188.
5. Hudeˇ
cek J, Voorwerk L, van Seijen M, et al. Application of a risk-
management framework for integration of stromal tumor-infiltrating
lymphocytes in clinical trials. NPJ Breast Cancer 2020; 6: 15.
6. Leon-Ferre RA, Jonas SF, Salgado R, et al. Abstract PD9-05: stromal
tumor-infiltrating lymphocytes identify early-stage triple-negative
breast cancer patients with favorable outcomes at 10-year follow-up
in the absence of systemic therapy: a pooled analysis of 1835 patients.
Cancer Res 2023; 83: PD9-05.
7. Loi S, Drubay D, Adams S, et al. Tumor-infiltrating lymphocytes and
prognosis: a pooled individual patient analysis of early-stage triple-
negative breast cancers. J Clin Oncol 2019; 37: 559–569.
8. Liang H, Li H, Xie Z, et al. Quantitative multiplex immunofluores-
cence analysis identifies infiltrating PD1
+
CD8
+
and CD8
+
T cells as
predictive of response to neoadjuvant chemotherapy in breast cancer.
Thorac Cancer 2020; 11: 2941–2954.
9. Russo L, Maltese A, Betancourt L, et al. Locally advanced breast
cancer: tumor-infiltrating lymphocytes as a predictive factor of
response to neoadjuvant chemotherapy. Eur J Surg Oncol 2019; 45:
963–968.
10. Morigi C. Highlights of the 16th St Gallen international breast cancer
conference, Vienna, Austria, 20–23 March 2019: personalised treat-
ments for patients with early breast cancer. Ecancermedicalscience
2019; 13: 924.
14 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

11. Danske Multidisciplinære Cancer Grupper. Patologiprocedurer og
molekylærpatologiske analyser ved brystkræft. Danske Multidisciplinære
Cancer Grupper: Copenhagen, Denmark. [Accessed 31 August 2021].
Available from: https://dmcg.dk.
12. Regionala Cancercentrum I Samverkan. Kvalitetsbilaga för bröstpatologi
(KVAST-bilaga). Kunskapsbanken. Regionala Cancercentrum I
Samverkan: Stockholm, Sweden. [Accessed 31 August 2021]. Available
from: https://kunskapsbanken.cancercentrum.se.
13. Salgado R, Denkert C, Demaria S, et al. The evaluation of tumor-
infiltrating lymphocytes (TILs) in breast cancer: recommendations by an
international TILs working group 2014. Ann Oncol 2015; 26: 259–271.
14. Kos Z, Roblin E, Kim RS, et al. Pitfalls in assessing stromal tumor
infiltrating lymphocytes (sTILs) in breast cancer. NPJ Breast Cancer
2020; 6: 17.
15. O’Loughlin M, Andreu X, Bianchi S, et al. Reproducibility and
predictive value of scoring stromal tumour infiltrating lymphocytes
in triple-negative breast cancer: a multi-institutional study. Breast
Cancer Res Treat 2018; 171: 1–9.
16. Kilmartin D, O’Loughlin M, Andreu X, et al. Intra-tumour heteroge-
neity is one of the main sources of inter-observer variation in scoring
stromal tumour infiltrating lymphocytes in triple negative breast can-
cer. Cancer 2021; 13: 4410.
17. van der Laak J, Litjens G, Ciompi F. Deep learning in histopathology:
the path to the clinic. Nat Med 2021; 27: 775–784.
18. Basavanhally AN, Ganesan S, Agner S, et al. Computerized image-based
detection and grading of lymphocytic infiltration in HER2+breast
cancer histopathology. IEEE Trans Biomed Eng 2010; 57: 642–653.
19. Yuan Y, Failmezger H, Rueda OM, et al. Quantitative image analysis
of cellular heterogeneity in breast tumors complements genomic
profiling. Sci Transl Med 2012; 4: 157ra143.
20. Amgad M, Sarkar A, Srinivas C, et al. Joint region and nucleus
segmentation for characterization of tumor infiltrating lymphocytes
in breast cancer. In Medical Imaging 2019: Digital Pathology,
Tomaszewski JE, Ward AD (eds). SPIE: San Diego, 2019; 20.
21. Saltz J, Gupta R, Hou L, et al. Spatial organization and molecular
correlation of tumor-infiltrating lymphocytes using deep learning on
pathology images. Cell Rep 2018; 23: 181–193.e7.
22. Thagaard J, Stovgaard ES, Vognsen LG, et al. Automated quantification
of sTIL density with H&E-based digital image analysis has prognostic
potential in triple-negative breast cancers. Cancers 2021; 13: 3050.
23. Bai Y, Cole K, Martinez-Morilla S, et al. An open-source, automated
tumor-infiltrating lymphocyte algorithm for prognosis in triple-
negative breast cancer. Clin Cancer Res 2021; 27: 5557–5565.
24. Sun P, He J, Chao X, et al. A computational tumor-infiltrating lympho-
cyte assessment method comparable with visualreporting guidelines for
triple-negative breast cancer. EBioMedicine 2021; 70: 103492.
25. Amgad M, Stovgaard ES, Balslev E, et al. Report on computational
assessment of tumor infiltrating lymphocytes from the international
immuno-oncology biomarker working group. NPJ Breast Cancer
2020; 6: 16.
26. Bankhead P, Loughrey MB, Fernández JA, et al. QuPath: open source
software for digital pathology image analysis. Sci Rep 2017; 7: 16878.
27. Le H, Gupta R, Hou L, et al. Utilizing automated breast cancer
detection to identify spatial distributions of tumor-infiltrating lympho-
cytes in invasive breast cancer. Am J Pathol 2020; 190: 1491–1504.
28. Abousamra S, Gupta R, Hou L, et al. Deep learning-based mapping of
tumor infiltrating lymphocytes in whole slide images of 23 types of
cancer. Front Oncol 2022; 11: 806603.
29. He T-F, Yost SE, Frankel PH, et al. Multi-panel immunofluorescence
analysis of tumor infiltrating lymphocytes in triple negative breast
cancer: evolution of tumor immune profiles and patient prognosis.
PLoS One 2020; 15: e0229955.
30. Swiderska-Chadaj Z, Pinckaers H, van Rijthoven M, et al. Learning to
detect lymphocytes in immunohistochemistry with deep learning.
Med Image Anal 2019; 58: 101547.
31. Balkenhol MCA, Ciompi F,
´
Swiderska-Chadaj _
Z, et al. Optimized
tumour infiltrating lymphocyte assessment for triple negative breast
cancer prognostics. Breast 2021; 56: 78–87.
32. Tellez D, Litjens G, Bándi P, et al. Quantifying the effects of data
augmentation and stain color normalization in convolutional neural net-
works for computational pathology. Med Image Anal 2019; 58: 101544.
33. Kohlberger T, Liu Y, Moran M, et al. Whole-slide image focus
quality: automatic assessment and impact on AI cancer detection.
J Pathol Inform 2019; 10: 39.
34. Smit G, Ciompi F, Cigéhn M, et al. Quality control of whole-slide
images through multi-class semantic segmentation of artifacts. In
MIDL 2021 Conference Short. Open Review: Amherst, MA, 2021.
35. Srinidhi CL, Ciga O, Martel AL. Deep neural network models for com-
putational histopathology: a survey. Med Image Anal 2021; 67: 101813.
36. Abe N, Matsumoto H, Takamatsu R, et al. Quantitative digital image
analysis of tumor-infiltrating lymphocytes in HER2-positive breast
cancer. Virchows Arch 2020; 476: 701–709.
37. Janowczyk A, Madabhushi A. Deep learning for digital pathology
image analysis: a comprehensive tutorial with selected use cases.
J Pathol Inform 2016; 7: 29.
38. Lu Z, Xu S, Shao W, et al. Deep-learning-based characterization of
tumor-infiltrating lymphocytes in breast cancers from histopathology
images and multiomics data. JCO Clin Cancer Inform 2020; 4: 480–490.
39. Chen J, Srinivas C. Automatic lymphocyte detection in H&E images
with deep neural networks. ArXiv preprint 2016; 1612.03217. [Not
peer reviewed].
40. Amgad M, Atteya LA, Hussein H, et al. NuCLS: A scalable
crowdsourcing approach and dataset for nucleus classification and
segmentation in breast cancer. GigaScience 2022; 11: giac037.
41. He K, Gkioxari G, Dollar P, et al. Mask R-CNN. IEEE Trans Pattern
Anal Mach Intell 2020; 42: 386–397.
42. Litjens G, Kooi T, Bejnordi BE, et al. A survey on deep learning in
medical image analysis. Med Image Anal 2017; 42: 60–88.
43. Swiderska-Chadaj Z, de Bel T, Blanchet L, et al. Impact of rescanning
and normalization on convolutional neural network performance in
multi-center, whole-slide classification of prostate cancer. Sci Rep
2020; 10: 14398.
44. Zarella MD, Bowman D, Aeffner F, et al. A practical guide to whole
slide imaging: a white paper from the digital pathology association.
Arch Pathol Lab Med 2019; 143: 222–234.
45. Abels E, Pantanowitz L, Aeffner F, et al. Computational pathology
definitions, best practices, and recommendations for regulatory guid-
ance: a white paper from the digital pathology association. J Pathol
2019; 249: 286–294.
46. de Bel T, Bokhorst J-M, van der Laak J, et al. Residual cyclegan for
robust domain transformation of histopathological tissue slides. Med
Image Anal 2021; 70: 102004.
47. Linmans J, van der Laak J, Litjens G. Efficient out-of-distribution
detection in digital pathology using multi-head convolutional neural
networks. Proc Mach Learn Res 2020; 121: 465–478.
48. Thagaard J, Hauberg S, van der Vegt B, et al. Can you trust predictive
uncertainty under real dataset shifts in digital pathology? In Medical
Image Computing and Computer Assisted Intervention –MICCAI
2020 (Vol. 12261. Lecture Notes in Computer Science.), Martel AL,
Abolmaesumi P, Stoyanov D, et al. (eds). Springer International
Publishing: Cham, 2020; 824–833.
49. Stacke K, Eilertsen G, Unger J, et al. A closer look at domain shift for
deep learning in histopathology. arXiv 2019; 1909.11575. [Not peer
reviewed].
50. Dudgeon SN, Wen S, Hanna MG, et al. A pathologist-annotated
dataset for validating artificial intelligence: a project description and
pilot study. J Pathol Inform 2021; 12: 45.
51. Amgad M, Elfandy H, Hussein H, et al. Structured crowdsourcing
enables convolutional segmentation of histology images.
Bioinformatics 2019; 35: 3461–3467.
Pitfalls in ML assessment of TILs 15
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

52. Wahab N, Miligy IM, Dodd K, et al. Semantic annotation for com-
putational pathology: multidisciplinary experience and best practice
recommendations. J Pathol Clin Res 2022; 8: 116–128.
53. Creative Commons —CC0 1.0 Universal. Creative Commons,
[Accessed 31 August 2021]. Available from: https://creativecommons.
org/publicdomain/zero/1.0/.
54. Creative Commons —Attribution-NonCommercial-ShareAlike 4.0
International —CC BY-NC-SA 4.0. Creative Commons, [Accessed
31 August 2021]. Available from: https://creativecommons.org/
licenses/by-nc-sa/4.0/.
55. Litjens G, Bandi P, Ehteshami Bejnordi B, et al. 1399 H&E-stained
sentinel lymph node sections of breast cancer patients: the
CAMELYON dataset. GigaScience 2018; 7: giy065.
56. Prostate cANcer graDe Assessment (PANDA) Challenge, Kaggle.
[Accessed 31 August 2021]. Available from: https://www.kagle.com.
57. Grand Challenge. Grand Challenge, [Accessed 31 August 2021].
Available from: https://grand-challenge.org/.
58. NIH. The Cancer Genome Atlas Program (TCGA). NIH National
Cancer Institute: Center for Cancer Genomics. [Accessed 31 August
2021]. Available from: https://www.cancer.gov/tcga.
59. Howard FM, Dolezal J, Kochanny S, et al. The impact of site-specific
digital histology signatures on deep learning model accuracy and bias.
Nat Commun 2021; 12: 4423.
60. U.S. Food & Drug Administration. Qualification of Medical Device
Development Tools, November 2013. [Accessed 31 August 2021].
Available from: https://www.fda.gov.
61. Kleppe A, Skrede O-J, De Raedt S, et al. Designing deep learning
studies in cancer diagnostics. Nat Rev Cancer 2021; 21:
199–211.
62. Acs B, Salgado R, Hartman J. What do we still need to learn on
digitally assessed biomarkers? EBioMedicine 2021; 70: 103520.
63. Stanton SE, Disis ML. Clinical significance of tumor-infiltrating
lymphocytes in breast cancer. J Immunother Cancer 2016;
4: 59.
64. Stanton SE, Adams S, Disis ML. Variation in the incidence and
magnitude of tumor-infiltrating lymphocytes in breast cancer sub-
types: a systematic review. JAMA Oncol 2016; 2: 1354.
65. Fassler DJ, Torre-Healy LA, Gupta R, et al. Spatial characterization of
tumor-infiltrating lymphocytes and breast cancer progression. Cancer
2022; 14: 2148.
16 J Thagaard, G Broeckx et al
© 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd
on behalf of The Pathological Society of Great Britain and Ireland. www.pathsoc.org
J Pathol 2023
www.thejournalofpathology.com
10969896, 0, Downloaded from https://pathsocjournals.onlinelibrary.wiley.com/doi/10.1002/path.6155 by Shamim Mushtaq - INASP/HINARI - PAKISTAN , Wiley Online Library on [24/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License