ArticlePDF Available

CT texture analysis: A potential tool for prediction of survival in patients with metastatic clear cell carcinoma treated with sunitinib

Authors:

Abstract and Figures

Background To assess CT texture based quantitative imaging biomarkers in the prediction of progression free survival (PFS) and overall survival (OS) in patients with clear cell renal cell carcinoma undergoing treatment with Sunitinib. Methods In this retrospective study, measurable lesions of 40 patients were selected based on RECIST criteria on standard contrast enhanced CT before and 2 months after treatment with Sunitinib. CT Texture analysis was performed using TexRAD research software (TexRAD Ltd, Cambridge, UK). Using a Cox regression model, correlation of texture parameters with measured time to progression and overall survival were assessed. Evaluation of combined International Metastatic Renal-Cell Carcinoma Database Consortium Model (IMDC) score with texture parameters was also performed. ResultsSize normalized standard deviation (nSD) alone at baseline and follow-up after treatment was a predictor of OS (Hazard ratio (HR) = 0.01 and 0.02; 95% confidence intervals (CI): 0.00 – 0.29 and 0.00 – 0.39; p = 0.01 and 0.01). Entropy following treatment and entropy change before and after treatment were both significant predictors of OS (HR = 2.68 and 87.77; 95% CI = 1.14 – 6.29 and 1.26 – 6115.69; p = 0.02 and p = 0.04). nSD was also a predictor of PFS at baseline and follow-up (HR = 0.01 and 0.01: 95% CI: 0.00 – 0.31 and 0.001 – 0.22; p = 0.01 and p = 0.003). When nSD at baseline or at follow-up was combined with IMDC, it improved the association with OS and PFS compared to IMDC alone. Conclusion Size normalized standard deviation from CT at baseline and follow-up scans is correlated with OS and PFS in clear cell renal cell carcinoma treated with Sunitinib.
Content may be subject to copyright.
R E S E A R C H A R T I C L E Open Access
CT texture analysis: a potential tool for
prediction of survival in patients with
metastatic clear cell carcinoma treated with
sunitinib
Masoom A. Haider
1*
, Alireza Vosough
2
, Farzad Khalvati
1
, Alexander Kiss
3
, Balaji Ganeshan
4
and Georg A. Bjarnason
5
Abstract
Background: To assess CT texture based quantitative imaging biomarkers in the prediction of progression free survival
(PFS) and overall survival (OS) in patients with clear cell renal cell carcinoma undergoing treatment with Sunitinib.
Methods: In this retrospective study, measurable lesions of 40 patients were selected based on RECIST criteria on
standard contrast enhanced CT before and 2 months after treatment with Sunitinib. CT Texture analysis was performed
using TexRAD research software (TexRAD Ltd, Cambridge, UK). Using a Cox regression model, correlation of texture
parameters with measured time to progression and overall survival were assessed. Evaluation of combined International
Metastatic Renal-Cell Carcinoma Database Consortium Model (IMDC) score with texture parameters was also performed.
Results: Size normalized standard deviation (nSD) alone at baseline and follow-up after treatment was a predictor of OS
(Hazard ratio (HR) = 0.01 and 0.02; 95% confidence intervals (CI): 0.00 0.29 and 0.00 0.39; p= 0.01 and 0.01). Entropy
following treatment and entropy change before and after treatment were both significant predictors of OS (HR = 2.68
and 87.77; 95% CI = 1.14 6.29 and 1.26 6115.69; p=0.02 and p= 0.04). nSD was also a predictor of PFS at baseline and
follow-up (HR = 0.01 and 0.01: 95% CI: 0.00 0.31 and 0.001 0.22; p=0.01and p= 0.003). When nSD at baseline or at
follow-up was combined with IMDC, it improved the association with OS and PFS compared to IMDC alone.
Conclusion: Size normalized standard deviation from CT at baseline and follow-up scans is correlated with OS and PFS in
clear cell renal cell carcinoma treated with Sunitinib.
Keywords: Prediction of outcome, Metastatic clear cell carcinoma, Quantitative imaging biomarkers, CT image features,
CT texture analysis
Background
Multi-targeted tyrosine kinase inhibitor (TKI) therapy
with Sunitinib is a standard treatment of metastatic clear
cell renal cell carcinoma (RCC). Non-imaging related
clinical prognostic factors have been identified for patients
receiving targeted therapy and introduced into treatment
guidelines and used to stratify patients on clinical trials
[1]. This clinical prognostic model and the associated
factors are described in the International Metastatic Renal-
Cell Carcinoma Database Consortium Model (IMDC),
which is used most commonly [2].
It is well known that enhancement features of RCC can
change on contrast enhanced CT in patients receiving
TKIs such as Sunitinib and this is not always reflected in
an early change in the size of tumors thus limiting the
application of RECIST criteria [3]. Multiple alternative
response criteria which combine size and enhancement
change such as Choi, modified Choi and Morphology
Attenuation, Size and Structure (MASS) criteria have
demonstratedapredictiveabilitybycombiningsizeand
enhancement criteria to predict progression free survival
(PFS) in patients with metastatic RCC [48].
* Correspondence: masoom.haider@sunnybrook.ca
1
Department of Medical Imaging, Sunnybrook Health Sciences Center,
University of Toronto, Rm AG-46, 2075 Bayview Ave, Toronto, Onatrio M4N
3M5, Canada
Full list of author information is available at the end of the article
© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Haider et al. Cancer Imaging (2017) 17:4
DOI 10.1186/s40644-017-0106-8
Intratumoral heterogeneity is a recognized feature of
cancer behavior and in particular, therapeutic resistance
[9]. Analysis of tumor heterogeneity using CT texture
analysis has shown promise as a prognostic and predict-
ive measure in RCC. A previous study by Goh et al.
showed that CT texture analysis reflecting tumor hetero-
geneity is an independent factor associated with PFS and
has the potential to be used as a predictive imaging bio-
marker of response of metastatic RCC of various histo-
logic types [10]. A more recent study has also included a
variety of RCC histologies and confirmed potential prog-
nostic value of CT texture features in assessment of the
primary tumor site and outcome [11]. There have been
other studies on the prognostic value of CT texture
features for different types of cancer including breast
[12] lung [13], hepatic metastatic colorectal cancer [14],
pancreatic cancer [15] as well as reproducibility of CT
texture parameters [16]. To our knowledge, a study
specifically reviewing the potential prognostic and
predictive value of texture features in a pure clear cell
RCC cohort and evaluating this in the light of the IMDC
prognostic score has not been performed.
The purpose of this study was to assess CT texture ana-
lysis based Quantitative Imaging Biomarkers (QIBs) in the
prediction of PFS and Overall Survival (OS) in patients
with clear cell RCC undergoing treatment with Sunitinib.
Methods
Patients
The institutional research ethics board approved this
retrospective single institution study and waived the
requirement for informed consent.
Patients with metastatic clear cell carcinoma who
received the TKI Sunitinib as first or second line therapy
at our institution between December 2005 and March
2010 were identified from institutional renal cancer
database. An attempt was made to optimize the activity
of Sunitinib by treating each patient to toxicity using
individualized dose and schedule [17]. An IMDC prog-
nostic score was assigned to each patient at baseline.
Patients were included if they were TKI naïve and had
received Sunitinib as the first or second line treatment
for metastatic clear cell carcinoma. Patients were ex-
cluded if: their baseline contrast enhanced CT was not
performed within 6 weeks before the start of treatment;
both their baseline and followup CT were not performed
with contrast enhancement; they did not have measur-
able disease at baseline as defined by RECIST 1.1.
From a total of 172 patients who were identified in the
institution database, 132 patients were excluded with 40
patients left for analysis. The most common reasons for
exclusion was lack of availability for the pre-treatment
or first follow-up scan, typically due to imaging being
performed at another institution.
CT examination
All patients underwent contrast enhanced CT examin-
ation of their chest, abdomen and pelvis (GE Lightspeed
Plus or GE Lightspeed VCT) following injection of
100 ml of an iodinated contrast agent (Omnipaque 300,
Iohexol, GE Healthcare, Princeton, NJ, USA) at a rate of
3 ml/s via an automated injector.
Images of the thorax were analyzed in arterial phase
(25-s delay) and images of the abdomen and pelvis were
analyzed in portal venous phase (70-s delay) with the
following acquisition parameters: 120 kV; auto mA and
Smart mA (angular and z-axis modulation); pitch 0.75:1
and 0.9:1; 20 mm collimation, 5 mm slice thickness and
40 mm collimation reconstructed; scan field of view
(FOV) 50 cm and display FOV adjusted to patient size;
matrix: 512x512 (pixel spacing: 0.933 mm). Region of
interest (ROI) was drawn by a radiologist with 6 years of
experience of reporting abdominal CT who was blinded
to the clinical outcome. Each ROI was drawn on the
slice through the largest diameter of the tumor site.
CT Texture Analysis
Target lesions were selected according to RECIST, version
1.1 (maximum of five target lesions, maximum of two le-
sions per organ). If the patient had not had resection of
primary tumor, it was included as a measurable tumor
and used for analysis. RECIST 1.1 criteria were used to
identify disease progression, which was confirmed using
subsequent imaging. Using RECIST 1.1, initial response
after two cycles of treatment was evaluated. CT texture
analysis of the lesions was performed using TexRAD com-
mercial research software (TexRAD Ltd, www.texrad.com,
part of Feedback Plc, Cambridge, UK) by drawing a ROI
around the peripheral margin of the selected lesions used
for RECIST. Metastases/primary lesions less than 1 cm in
maximal diameter were not included in the analysis. Air,
streak artifacts and dense calcifications were excluded
from the regions of interest. CT texture analysis com-
prised a filtration-histogram technique where the filtration
step extracted and enhanced (amplified) features using a
band-pass Laplacian of Gaussian spatial scale filter [18].
Quantification was done using different histogram based
statistical parameters in the selected region of interest in
CT images after the application of the band-pass filter at
intermediate scale which was chosen to mimic the scale
used by Goh et al. [10].
Histogram based statistical parameters comprised of
mean positive pixel intensity (the average value of the
positive pixels within the ROI), standard deviation (SD),
skewness (symmetry of the pixel intensity distribution),
kurtosis (pointiness of the pixel intensity distribution),
and entropy which represents irregularity or complexity
of pixel intensity in space. These are all first-order statis-
tical features except for entropy which is a second-order
Haider et al. Cancer Imaging (2017) 17:4 Page 2 of 9
statistical feature. A newer secondary histogram param-
eter derived from the above parameters, size normalized
standard deviation (nSD), was also evaluated. This was
done as it is known that SD estimate can be affected by
the size (meaning the number of pixels) specifically in
case of small tumors which may need correction or
normalization [18] (Eq. 1).
nSD of a given ROI is calculated as follows:
nSD ¼Ln SDðÞ
Ln NðÞ ð1Þ
where SD is the standard deviation of ROI (i.e., tumor)
and Nis the total number of pixels in the ROI.
These parameters were recorded for each lesion in
baseline CT images and the first follow-up CT images
after treatment with Sunitinib. Percentage change (PerC)
of the CT texture parameters before and after treatment
were calculated (Eq. 2).
PerC ¼Parameter post treatmentðÞParameter pretreatmentðÞ
Parameter pretreatment
ðÞ
100
ð2Þ
As the value of Kurtosis can vary from 3to+3,to
negate division by 0 when calculating the percentage
change, we added 3, i.e. [((Kurtosis post treatment +3) -
(Kurtosis pre-treatment +3))/(Kurtosis pre-treatment +3)]
× 100. In case of Skewness, in order to avoid dividing by
0, only change (not the percentage change) was used, i.e.
Skewness post treatment Skewness pre-treatment.
Statistical Analysis
For statistical analysis, average texture measurements of
all measured lesions in each patient at baseline CT and
the first follow-up CT performed about 2 months after
the start of treatment, the percentage change from
baseline value as well as the percentage change in lesion
size based on RECIST 1.1 criteria, which measures the
amount of lesion size reduction, were used. PFS was
defined as the time from the date of baseline CT to the
date of disease progression based on RECIST response
criteria. A Cox proportional hazards survival model was
performed to determine if any parameter was predictors
of OS or PFS. No adjustments for multiple testing were
carried out as these analyses are considered exploratory
and their results will serve to enhance future larger
studies. All statistical analysis was performed using SAS
Version 9.3 (SAS Institute, Cary, NC, USA).
The International Metastatic Renal-Cell Carcinoma
Database Consortium Model (IMDC) allocates patient
to three prognostic groups (good, intermediate and
poor) based on the degree of anemia, thrombocytosis,
neutrophilia, and hypercalcemia, as well as the Karnofski
performance status <80%, and <1 year from diagnosis to
treatment. To assess whether any of the CT texture
parameters adds to IMDC in prediction of PFS and OS,
we ran 2 variable Cox proportional hazards regression
models and assessed model fit using 2 log likelihood
(2LL) statistics. For addition of one variable to IMDC,
a change of greater than 3.84 (the critical value associ-
ated with one degree of freedom for a chi-square statis-
tic) would indicate a significant model improvement.
We could not explore models with 3 or more variables
due to limitations in sample size. A P value of less than
0.05 was considered to indicate statistical significance.
Results
The cohort consisted of 35 men and 5 women with a
mean age of 60 years (range of 3476 years). All the
patients had clear cell carcinoma. 34 patients received
Sunitinib as first line therapy and 6 patients received
Sunitinib as the second line treatment after being treated
with a non-TKI drug (e.g. interferon). A total of 87
target lesions were analyzed. These lesions were scat-
tered across different locations as listed in Table 1.
The baseline contrast enhanced CT of the patients in-
cluded in the analysis was performed within 6 weeks before
the start of treatment (mean was 12.8 day pre-treatment)
and their followup CT was performed with contrast
enhancement (mean followup time was 67.5 days). Six of
40 patients had progressed during the study period with a
PFS of 60 days (range: 64171 days). The length of follow-
up was up to 22.8 months following baseline CT or until
death whichever sooner. The significance of each texture
Table 1 Tumor Sites
Tumor Site Number of lesions
Lymph node 15
Lung 14
Primary 10
Bone 10
Liver 8
Peritoneum 6
Adrenal 4
Mediastinal Lymph Nodes 4
Nephrectomy bed 4
Other kidney 3
Hilar Lymph Nodes 3
Pleura 1
IVC Thrombus 1
Muscle 1
Spleen 1
Omentum 1
Psoas major muscle 1
Total 87
Haider et al. Cancer Imaging (2017) 17:4 Page 3 of 9
parameter in prediction of OS and PFS is summarized in
Tables 2 and 3, respectively.
Overall survival
Size normalized standard deviation (nSD) prior to treat-
ment with Sunitinib was a significant predictor for OS
(p= 0.01) such that higher nSD before treatment pre-
dicted for increased survival (positive correlation) (HR
= 0.01, 95% CI = 0.00 0.29).Inaddition,nSDfollow-
ing treatment with Sunitinib was found to be significant
in predicting OS (p= 0.01) such that higher nSD after
treatment was associated with a lower hazard of death
or higher survival (positive correlation) (HR = 0.02, 95%
CI = 0.001 0.39). Entropy following treatment and en-
tropy change both were significant predictors of OS
such that higher entropy following treatment or en-
tropy change predicted decreased survival (negative
correlation) (p=0.02, HR=2.68, 95% CI=1.14 6.29
and p= 0.04, HR = 87.77, 95% CI = 1.26 6115.69,
respectively) (Table 2).
Progression free survival
With regards to PFS, nSD prior to treatment with Sunitinib
was a significant predictor (p= 0.01) such that higher nSD
before treatment was related to increased PFS (positive
correlation) (HR = 0.01, 95% CI = 0.00 0.31). Moreover,
nSD following treatment with Sunitinib was found to be
significant in predicting PFS (p= 0.003) such that higher
nSD after treatment leads to lower hazard of death or in-
creased PFS (positive correlation) (HR = 0.01, 95% CI = 0.00
0.22). Neither entropy following treatment nor entropy
change was significant predictor of PFS (Table 3).
The lesions size change (percentage change, in terms
of reduction, in sum of the lesionssize in each patient
prior to and following treatment with Sunitinib) demon-
strated a significant relation to OS (positive correlation)
(p= 0.02), (HR = 0.97, 95% CI = 0.94 1.00), while size
change was not a significant predictor of PFS.
Imaging variables combined with IMDC
To analyze the added predictive value of imaging vari-
ables to IMDC, we started with a baseline Cox propor-
tion hazards model that only included the IMDC and
then added each of the imaging variables found to be
significant in the univariate survival analyses for OS or
PFS. The baseline model with IMDC alone was signifi-
cantly related to OS and had a 2 log likelihood statistic
of 181.3. Therefore, to see an improved fit in any of the
two-variable models tested, this statistic would have to
change by a minimum of 3.84 (a 1° of freedom change
Table 2 Imaging parameters as predictors of OS
Variable P-value Hazard ratio 95% CI
Mean positive pixel intensity prior
to treatment
0.34 1.01 0.98 1.05
Mean positive pixel intensity
following treatment
0.77 1.00 0.99 1.02
SD prior to treatment 0.24 0.98 0.96 1.01
SD following treatment 0.46 0.99 0.98 1.01
Entropy prior to treatment 0.27 1.74 0.65 4.68
Entropy following treatment 0.02 2.68 1.14 6.29
nSD prior to treatment 0.01 0.01 0.00 0.29
nSD following treatment 0.01 0.02 0.001 0.39
Kurtosis prior to treatment 0.26 1.24 0.85 1.82
Kurtosis following treatment 0.78 1.03 0.81 1.32
Skewness prior to treatment 0.15 2.02 0.77 5.30
Skewness following treatment 0.30 1.47 0.70 3.09
Percent change in size 0.02 0.97 0.94 0.99
Mean pixel intensity Change 0.37 0.99 0.98 1.01
SD Change 0.80 1.14 0.42 3.09
Entropy Change 0.04 87.77 1.26 6115.69
nSD Change 0.39 0.18 0.004 8.56
Kurtosis Change 0.46 0.66 0.21 2.02
Skewness Change 0.98 0.99 0.60 1.65
Entries in bold were significant
Abbreviations: SD standard deviation, nSD size normalized standard deviation
OS overall survival
Table 3 Imaging parameters as predictors of PFS
Variable P-value Hazard ratio 95% CI
Mean positive pixel intensity
prior to treatment
0.61 1.01 0.98 1.04
Mean positive pixel intensity
following treatment
0.72 1.00 0.99 1.02
SD prior to treatment 0.71 0.99 0.97 1.02
SD following treatment 0.33 0.99 0.97 1.01
Entropy prior to treatment 0.25 1.84 0.64 5.32
Entropy following treatment 0.08 2.23 0.91 5.49
nSD prior to treatment 0.01 0.01 0.000 0.31
nSD following treatment 0.003 0.01 0.001 0.22
Kurtosis prior to treatment 0.21 1.28 0.87 1.87
Kurtosis following treatment 0.34 1.12 0.89 1.41
Skewness prior to treatment 0.32 1.66 0.62 4.41
Skewness following treatment 0.65 1.20 0.55 2.61
Percent change in size 0.14 0.98 0.96 1.01
Mean pixel intensity change 0.36 0.99 0.98 1.01
SD change 0.51 0.70 0.25 2.02
Entropy change 0.21 15.12 0.21 1096.9
nSD change 0.06 0.02 0.00 1.21
Kurtosis change 0.92 1.06 0.37 3.04
Skewness change 0.81 0.94 0.56 1.58
Entries in bold were significant
Abbreviations: SD standard deviation, nSD size normalized standard deviation,
PFS progression free survival
Haider et al. Cancer Imaging (2017) 17:4 Page 4 of 9
for a chi-square statistic). This means that any value of
177.5 or less indicated a significant improvement in
model fit. This improvement was seen for all two-
variable models tested, however the ones that contained
nSD prior to treatment and nSD following treatment
showed the greatest improvement in fit. For both of
these models, IMDC was no longer significant. There-
fore, these variables (i.e., nSD prior to and following
treatment) were seen as stronger predictors of survival
in a model that included IMDC. The ability of each
texture parameter to add to IMDC model in the predic-
tion of OS is summarized in Table 4.
For PFS models, the baseline model with IMDC
alone was not significantly related to PFS and had a
2 log likelihood statistic of 174.4. Therefore to see an
improved fit in any of the two-variable models tested,
this statistic would have to change by a minimum of
3.84 (a 1° of freedom change for a chi-square statistic).
This means that any value of 170.6 or less indicated a
significant improvement in model fit. This improve-
ment was only seen for the 2 variable models that con-
tained nSD prior to treatment and nSD following
treatment. For both of these models, the nSD variables
significantly improved the model in predicting PFS
whereas in neither model IMDC was significant.
Therefore, these variables were seen as stronger pre-
dictors of survival in a model that included IMDC.
The ability of each texture parameter to improve
IMDC model in the prediction of PFS is summarized
in Table 5.
When added, the lesion size (percentage change in
sum of the largest diameter of the lesionssize prior to
and post treatment) did not make a significant
improvement to IMDC in prediction of OS. Finally,
separate univariate Cox proportional hazards models
wererun,oneforpercentchangeinsize,theotherfor
nSD prior to treatment, to compare the two variables.
nSD prior to treatment was found to be a better pre-
dictor of OS (p= 0.01 versus 0.02) with HR = 0.01, 95%
CI = 0.00 0.29).
Examples of CT images demonstrating the appearance
of lesions in patients with varying outcomes (OS and PFS)
are presented in Fig. 1 and Table 6. The Kaplan-Meier
plots for significant parameters for OS and PFS are shown
in Figs. 2 and 3.
Discussion
In this study, we demonstrate that texture related CT
QIBs are predictive of OS and PFS in metastatic RCC
treated with Sunitinib. This study is different from prior
studies including the ones done by Goh et al. [10] and
Lubner et al. [11] in that the population is limited to pa-
tients with clear cell RCC and that it assesses the im-
aging features in combination with the IMDC
prognostic model for RCC patients treated with targeted
therapy [1]. This is clinically relevant as one of the pro-
posed uses of QIBs for personalized medicine is the se-
lection of optimal drug therapies. In the context of clear
cell RCC, there are more than one potential first line
TKIs. Thus, it is important to know the baseline charac-
teristic in the context of single histologic tumor type
with the standard first line agent as a monotherapy. We
have shown that size normalized standard deviation at
baseline and follow-up, tumor size change, entropy at
initial follow-up, and entropy change before and after
treatment are potentially useful QIBs.
In addition, the study in Goh et al. [10] only looks at
PFS while in this paper, both OS and PFS are studied.
Furthermore, we limited our cohort to clear cell carcin-
oma only and a single TKI drug while Goh et al. [10] in-
cludes multiple TKIs and multiple tumor types. Thus,
the cohort used in this study would be more reflective to
be used as a benchmark of baseline Sunitinib activity to
compare new TKIs in drug trials. The study by Lubner
et al. [11] includes a variety of RCC subtypes and does
Table 4 Imaging variables combined with IMDC to see if they
improve on the prediction of OS
Model IMDC IMDC Combined
Model
Fit statistic value
(2LL)
Baseline model
with IMDC alone
Significant 181.3
IMDC + Entropy
following treatment
Not Significant Significant 176.4
IMDC + nSD prior
to treatment
Not significant Significant 176.1
IMDC + nSD
following treatment
Not significant Significant 175.3
IMDC + percent
change in lesion size
Not significant Not significant 177.8
IMDC + Entropy
change
Not significant Not significant 177.4
Entries in bold were significant
Abbreviations: nSD size normalized standard deviation, IMDC International
Metastatic Renal-Cell Carcinoma Database Consortium Model, OS
overall survival
Table 5 Imaging variables combined with IMDC to see if they
improve on prediction of the PFS
Model IMDC IMDC Combined
Model
Fit statistic
value (2LL)
Baseline model with
IMDC alone
Not significant 174.4
IMDC + nSD
pre treatment
Not significant Significant 169.3
IMDC + nSD
post treatment
Not significant Significant 165.9
Entries in bold were significant
Abbreviations: nSD = size normalized standa rd deviation, IMDC International
Metastatic Renal-Cell Carcinoma Database Consortium Model, PFS progression
free survival
Haider et al. Cancer Imaging (2017) 17:4 Page 5 of 9
not specify what drugs used for patient treatment and
focuses more on prediction of histologic features.
CT texture features and IMDS
Given that the IMDC prognostic model is well estab-
lished in RCC patients, we have demonstrated the
potential additional value of nSD to further improve pre-
diction of OS and PFS when added to IMDC. In our
study, nSD measured in CT images acquired before and
after treatment were added to IMDC model. The
entropy change prior to and following the treatment and
percentage change in the sum of the largest diameter of
Fig. 1 Baseline CT in patients with metastatic clear cell renal cell carcinoma and intermediate IMDC score (Left: prior to treatment, Right: following the
treatment). a72-year-old man: OS and PFS in this patient was 1074 and 1063 days, respectively with a baseline nSD = 0.66. Although the tumor remains
heterogeneous in appearance there is response by RECIST criteria. b40-year-old man: OS and PFS in this patient were poor at 159 and 113 days, respectively
with a lower baseline nSD of 0.54. The tumor is slightly larger but stable by RECIST criteria. The list of significant parameters for 2 cases is shown in Table 6
below. As it can be seen from the table, higher nSD prior to and following treatment and higher percent change in lesion size (i.e., higher size reduction)
were associated with higher survival; and higher entropy following treatment and entropy change predicted decreased survival. It should be noted that for
PFS, only nSD prior to and following treatment were statistically significant parameters
Table 6 The list of significant parameters (with median values) for 2 cases shown in Fig. 1
OS PFS Entropy following treatment nSD prior to treatment nSD following treatment Percent change in size Entropy Change
Median 729 358 4.76 0.58 0.56 7.65 0.036
A 1074 1063 4.57 0.66 0.88 33 0.119
B 159 113 4.79 0.54 0.55 10.002
Haider et al. Cancer Imaging (2017) 17:4 Page 6 of 9
the lesions size prior to and post treatment did not make
a significant improvement to IMDC in the prediction of
OS. In a model comparing size and nSD prior to treat-
ment, the latter was a better predictor of OS.
CT texture features and RECIST criteria
The fact that early size change measured by RECIST is
correlated with OS but not PFS is interesting. This dis-
crepancy might be explained by the fact that PFS is re-
flective of the time course of a single drug as patients
would not switch to a second line TKI unless they pro-
gressed while OS includes a course of therapy which
may include second line and third line TKIs. Thus, the
early size change on Sunitinib could be reflective of
overall responsiveness to the TKIs family including sec-
ond and third line therapies. RECIST is a very intuitive
QIB. In contrast, the visual correlate of nSD may not be
intuitive and this may be a barrier to adoption. There-
fore, further study is required to investigate the visual
patterns of QIBs such as nSD and their histologic corre-
lates to make them more meaningful to the clinicians so
that they are not considered as the output of a black
box. A preliminary study has been done in this regard
by Miles et al. [18] however, further work is required.
Fig. 2 Kaplan-Meier plot of cumulative overall survival (OS) (a) Entropy following treatment (median: 4.76), (b) nSD prior to treatment (median: 0.58),
(c) nSD following treatment (median: 0.56), (d) Percent change in size (median: 7.65), and E) Entropy Change (median: 0.036)
Fig. 3 Kaplan-Meier plot of cumulative progression free survival (PFS) (a) nSD prior to treatment (median: 0.58) and (b) nSD following treatment
(median: 0.56)
Haider et al. Cancer Imaging (2017) 17:4 Page 7 of 9
For example, it is possible that nSD, by combining the
number of voxels in a ROI and standard deviation, is
combining both size and variation into a single bio-
marker and thus, providing value by not wholly
dispensing with size as a QIB.
Incorporation of new imaging response criteria such
as changes in attenuation, morphology and structure
into new classifications (Choi, modified Choi and
Morphology, Attenuation, Size, and Structure (MASS)
criteria) [4, 8] to assess response to antiangiogenic ther-
apies provides more accurate assessment of tumor
response to the targeted therapies, but has some limita-
tions. These criteria require follow-up imaging and are
not in themselves QIBs, however, they are intuitive and
pragmatic. Further research into the optimal approach
and use of these measures in personalized medicine is
required.
Study limitations
Our study has limitations. In particular, we note that the
small sample size did not allow for multiple testing cor-
rection for the large number of QIBs tested. This means
that the results for this study remain hypothesis generat-
ing and further prospective validation will be required as
with other similar studies published to date. In addition,
this study was retrospective, raising potential selection
biases. For example, patients who had a decline in renal
function and did not have contrast on their follow-up
were excluded from this analysis.
Clinical value
Of clinical interest is the value of baseline nSD both as a
predictive and prognostic parameter and its potential
additive value to the IMDC score at baseline. This opens
the door to prospective validation with the aim to define
a predictive biomarker for optimal drug selection ini-
tially without having to wait for a follow-up scan using a
practical and low cost standard of care test, a simple
contrast enhanced CT.
Conclusion
In conclusion, size normalized standard deviation is a
quantitative imaging biomarker texture feature that in
patients with metastatic clear cell carcinoma may add to
the IMDC score in prediction of therapy response and
overall survival. Further validation of this biomarker in
prospective trials is required.
Abbreviations
CI: Confidence interval; HR: Hazard ratio; IMDC: International Metastatic
Renal-Cell Carcinoma Database Consortium Model; nSD: normalized standard
deviation; OS: Overall survival; PerC: Percentage change; PFS: Progression free
survival; QIB: Quantitative imaging biomarker; RCC: Renal cell carcinoma;
ROI: Region of interest; SD: Standard deviation; TKI: Multi-targeted tyrosine
kinase inhibitor
Acknowledgements
Not applicable.
Funding
This research has been supported by Ontario Institute for Cancer Research
(OICR).
Availability of data and material
The datasets during and/or analyzed during the current study available from
the corresponding author on reasonable request pending the approval of
the institution and trial/study investigators who contributed to the dataset.
Authorscontributions
MAH, AV, FK, AK, BG, and GAB contributed to the design and
implementation of the concept. MAH, AV, and GAB contributed in collecting
and reviewing the data. MAH, FK, and AK contributed to the design and
implementation of quantitative feature extraction modules. AK contributed
to the statistical analysis of the data. All authors contributed to the writing
and reviewing of the paper. All authors read and approved the final
manuscript.
Competing interests
B. Ganeshan is a director, part-time employee and shareholder of Feedback
Plc (Cambridge, UK), a company that develops and commercializes the TexRAD
texture analysis research software analysis described in this manuscript.
The remaining authors declare that they have no competing interests.
Consent for publication
Not applicable.
Ethics approval and consent to participate
The institutional research ethics board approved this retrospective single
institution study and waived the requirement for informed consent.
Author details
1
Department of Medical Imaging, Sunnybrook Health Sciences Center,
University of Toronto, Rm AG-46, 2075 Bayview Ave, Toronto, Onatrio M4N
3M5, Canada.
2
Department of Radiology, North Bristol NHS Trust, Southmead
Hospital, Bristol, UK.
3
Sunnybrook Health Sciences Center, Toronto, Canada.
4
Institute of Nuclear Medicine, University College London, London, UK.
5
Sunnybrook Odette Cancer Centre, Division of medical Oncology, University
of Toronto, Toronto, ON, Canada.
Received: 16 September 2016 Accepted: 10 January 2017
References
1. National Comprehensive Cancer Network-Kidney Cancer 3. 2016.
2. Heng DYC, Xie W, Regan MM, Harshman LC, Bjarnason GA, Vaishampayan UN,
et al. External validation and comparison with other models of the international
metastatic renal-cell carcinoma database consortium prognostic model: a
population-based study. Lancet Oncol. 2013;14(2):1418.
3. Griffin N, Gore ME, Sohaib SA. Imaging in metastatic renal cell carcinoma.
AJR Am J Roentgenol. 2007;189(2):36070.
4. van der Veldt AAM, Meijerink MR, van den Eertwegh AJM, Haanen JBAG,
Boven E. Choi response criteria for early prediction of clinical outcome in
patients with metastatic renal cell cancer treated with sunitinib. Br J Cancer.
2010;102(5):8039.
5. Schmidt N, Hess V, Zumbrunn T, Rothermundt C, Bongartz G, Potthast S.
Choi response criteria for prediction of survival in patients with metastatic
renal cell carcinoma treated with anti-angiogenic therapies. Eur Radiol.
2013;23(3):6329.
6. Thian Y, Gutzeit A, Koh D-M, Fisher R, Lote H, Larkin J, et al. Revised choi
imaging criteria correlate with clinical outcomes in patients with metastatic
renal cell carcinoma treated with sunitinib. Radiology. 2014;273(2):132702.
7. Kinkel K, Helbich TH, Esserman LJ, Barclay J, Schwerin EH, Sickles EA, et al.
Dynamic high-spatial-resolution MR imaging of suspicious breast lesions:
diagnostic criteria and interobserver variability. AJR Am J Roentgenol.
2000;175(1):3543.
8. Smith AD, Shah SN, Rini BI, Lieber ML, Remer EM. Morphology, Attenuation,
Size, and Structure (MASS) criteria: Assessing response and predicting
Haider et al. Cancer Imaging (2017) 17:4 Page 8 of 9
clinical outcome in metastatic renal cell carcinoma on antiangiogenic
targeted therapy. Am J Roentgenol. 2010;194(6):14708.
9. Marusyk A, Almendro V, Polyak K. Intra-tumour heterogeneity: a looking
glass for cancer? Nat Rev Cancer. 2012;12(5):32334.
10. Goh V, Ganeshan B, Nathan P, Juttla JK, Vinayan A, Miles KA. Assessment of
response to tyrosine kinase inhibitors in metastatic renal cell cancer:
CT texture as a predictive biomarker. Radiology. 2011;261(1):16571.
11. Lubner MG, Stabo N, Abel EJ, Munoz Del Rio A, Pickhardt PJ. CT textural
analysis of large primary renal cell carcinomas: pretreatment tumor
heterogeneity correlates with histologic findings and clinical outcomes.
Am J Roentgenol. 2016;207(1):96105.
12. Zhu Y, Li H, Guo W, Drukker K, Lan L, Giger ML, et al. Deciphering genomic
underpinnings of quantitative MRI-based radiomic phenotypes of invasive
breast carcinoma. Nat Sci Reports. 2015;5:17787.
13. Coroller TP, Grossmann P, Hou Y, Rios Velazquez E, Leijenaar RTH, Hermann G,
et al. CT based radiomic signature CT-based radiomic signature predicts distant
metastasis in lung adenocarcinoma. Radiother Oncol. 2015. [cited 2016 Apr
29];114:34550. Available from: http://dx.doi.org/ 10.1016/j.radonc.2015.02.015.
14. Lubner MG, Stabo N, Lubner SJ, del Rio AM, Song C, Halberg RB, et al.
CT textural analysis of hepatic metastatic colorectal cancer: pre-treatment
tumor heterogeneity correlates with pathology and clinical outcomes.
Abdom Imaging. Springer US; 2015 Oct 13 [cited 2016 May 3];40(7):23317.
Available from: http://link.springer.com/10.1007/s00261-015-0438-4.
15. Campbell D. CT Textural Analysis (CTTA) of Metastatic Treatmenti-Resistant
Pancreatic Adenocarcinoma (PDAC): Identifying Biomarkers for Genetic
Instability and Overall Survival [Internet]. The University of Arizona College
of Medicine; 2016 [cited 2017 Jan 5]. Available from: http://hdl.handle.net/
10150/603564.
16. Zhao B, Tan Y, Tsai W-Y, Qi J, Xie C, Lu L, et al. Reproducibility of radiomics for
deciphering tumor phenotype with imaging. Nat Sci Reports [Internet]. 2016
[cited 2016 Apr 29];6:23428. Available from: http://doi.org/10.1038/srep23428.
17. Bjarnason GA, Khalil B, Hudson JM, Williams R, Milot LM, Atri M, et al. Outcomes
in patients with metastatic renal cell cancer treated with individualized sunitinib
therapy: correlation with dynamic microbubble ultrasound data and review of
the literature. Urol Oncol Semin Orig Investig. 2014;32(4):4807. Available from:
http://dx.doi.org/10.1016/j.urolonc.2015.03.003.
18. Miles KA, Ganeshan B, Hayball MP. CT texture analysis using the filtration-
histogram method: what do the measurements mean? Cancer Imaging.
2013;13(3):4006.
We accept pre-submission inquiries
Our selector tool helps you to find the most relevant journal
We provide round the clock customer support
Convenient online submission
Thorough peer review
Inclusion in PubMed and all major indexing services
Maximum visibility for your research
Submit your manuscript at
www.biomedcentral.com/submit
Submit your next manuscript to BioMed Central
and we will help you at every step:
Haider et al. Cancer Imaging (2017) 17:4 Page 9 of 9
... Scarce data is currently available about CT-based radiomics for predicting response to different treatments. Mühlbauer reported only six studies investigating quantitative imaging for predicting response to systemic therapy 21 , of which only two applied CT-based radiomics to predict response to anti-angiogenic treatment with limited data on radiomics model accuracy 37,38 . Both articles implemented a univariate radiomics Cox regression model rather than combining different radiomics features in a multivariate predictive model. ...
... However, both studies analyzed changes between baseline and after administration of tyrosine kinase inhibitors (TKIs). Haider 38 found two radiomics features (entropy and size normalized standard deviation) significantly associated with survival at baseline. Goh 37 found that texture uniformity at baseline could predict the time to progression. ...
Article
Full-text available
The application of advanced computational analysis to medical imaging opens a plethora of opportunities in the field of radiology, allowing for more accurate tissue characterization and, eventually, advancing towards precision medicine through imaging biomarkers. In this review, we briefly introduce the methodology for radiomics analysis and the main challenges for implementation of radiomics-based tools in clinical practice. Based on systematic review of published studies, we also summarize here the main advances regarding CT-based radiomics applications in renal cancer with regards to tumor characterization (diagnosis, grading, prognosis), gene expression prediction (radiogenomics) and response evaluation.
... Regarding kidney tumors, the Radiomic texture analysis showed an area under the curve (AUC) of 0.80 to discriminate between renal cell carcinoma and papillary carcinoma (19). In a recent study, radiomic texture analysis parameters Entropy (ENT) and standard deviation (SD) showed a correlation with overall survival in clear cell carcinoma treated with Sunitinib (20). ...
Article
Full-text available
Renal cancer has a global incidence and mortality of 2.2 and 1.8%, respectively. Up to 30% of these patients are intrinsically resistant to tyrosine kinase inhibitors (TKI). The National Comprehensive Cancer Network guidelines do not include any predictive factors regarding response to systemic therapy with TKI in recurrent and advanced diseases. The present study aimed to explore whether a model based on radiomics could predict treatment response in patients with advanced kidney cancer treated with TKIs. The current study included 62 patients with advanced kidney cancer (stages 3 and 4) that underwent a CT scan in the arterial phase from March 2016 to November 2020. Texture analysis was run on the largest cross-sectional area of the primary tumor from each CT scan. A total of three different models were built from radiomics features and clinical data to analyze them by logistic regression and determine whether they correlated with the response to TKI. A receiver operating characteristic curve analysis was performed in each model to calculate the area under the curve (AUC) and the 95% confidence interval (CI). Significant radiomics features and clinical variables were identified and then a clinical model was created (AUC=0.90; sensitivity 75%; specificity 82.35%; CI 95%, 0.78-1.00), a radiomic model (AUC=0.66; sensitivity 16.67%; specificity 89.47%, CI 95%, 0.45-0.87) and a combined model (AUC=0.94; sensitivity 83.33%; specificity 94.12%; CI 95%, 0.84-1.00). Overall, models based on clinical data and radiomics could anticipate response to systemic therapy with TKI in patients with advanced kidney cancer.
... At present, the treatment options for RCC patients are very limited besides surgery [4]. It is reported that targeted biologics, such as tyrosine kinase inhibitor [5] and rapamycin inhibitor [6], have exhibited certain clinical efficacy for RCC. Nevertheless, these mentioned biologics easily produce drug resis-tance and negative adverse effects [7]. ...
Article
Full-text available
Background: Renal cell carcinoma (RCC) is a frequent disease with limited curative methods. This study is aimed at investigating the role and mechanism of Radix Actinidia chinensis (RAC) on RCC. Methods: The ingredients, target, and crucial pathways of RAC in RCC therapy were analyzed by network pharmacology. Then, an RCC animal model was established by subcutaneously injecting A498 cell suspension to BALB/c nude mice. After 1 week, the mice in the RAC-L/M/H groups were administered with RAC at 5, 10, and 20 mg/kg/d, respectively. The histopathology of the tumor was evaluated. The contents of tumor inflammatory cytokines and serum oxidative stress factors were detected by ELISA. The apoptosis of tumor tissues was assessed by TUNEL staining. The expressions of apoptosis-, proliferate-, autophagy-, and MAPK-related proteins were measured. Results: There were 13 active ingredients, and 20 RCC-relevant targets were selected from RAC; KEGG pathway indicated that these targets were enriched in the PI3K/AKT/mTOR and MAPK pathway. In in vivo experiments, RAC not only obviously damaged tumor cells and decreased the release of inflammatory cytokines and oxidative stress factors but also enhanced the apoptosis of the tumor cell in RCC mice. Besides, the expressions of apoptosis-, proliferate-, autophagy-, PI3K/AKT/mTOR path-, and MAPK path-related proteins were all affected by RAC. Conclusion: RAC attenuated RCC by regulating inflammation response, oxidative stress, apoptosis, proliferation, and autophagy, and its effects were partly linked to the PI3K/AKT/mTOR and MAPK pathway, which indicated that RAC may be a candidate drug for RCC.
... [15][16][17][18][19] There have been conflicting results on the direction of relationship between kurtosis and prognosis of patients with cancer; several studies showed that lower kurtosis were related with shorter OS, 52,53 as in our study, but others reported higher kurtosis in patients with shorter OS and PFS. 54,55 More studies are needed to further clarify the association between kurtosis and outcome. ...
Article
Full-text available
Background: The treatment responses of immune checkpoint inhibitors in metastatic renal cell carcinoma (mRCC) vary, requiring reliable prognostic biomarkers. We assessed the prognostic ability of computed tomography (CT) texture analysis in patients with mRCC treated with programmed death receptor-1 (PD-1)/programmed death ligand-1 (PD-L1) inhibitors. Materials and methods: Sixty-eight patients with mRCC treated with PD-1/PD-L1 inhibitors between 2012 and 2019 were revaluated. Using baseline and first follow-up CT, baseline and follow-up texture models were developed to predict overall survival (OS) and progression-free survival (PFS) using least absolute shrinkage and selection operator Cox-proportional hazards analysis. Patients were divided into high-risk or low-risk group, and the survival difference was assessed using Kaplan-Meier and log-rank test. Multivariable Cox models were constructed by including only the clinical variables (clinical models) and by combining the clinical variables and the texture models (combined clinical-texture models), and their predictive performance was evaluated using Harrell's C-index. Results: The baseline texture models distinguished longer- and shorter-term survivors for both OS (median, 60.1 vs. 17.0 months; P = .048) and PFS (5.2 vs. 2.8 months; P = .003). The follow-up texture models distinguished longer- and shorter-term overall survivors (40.3 vs. 15.2 months; P = .008) but not for PFS (5.0 vs. 3.6 months; P = .25). The combined clinical-texture model outperformed the clinical model in both predicting the OS (C-index, 0.70 vs. 0.63; P = .03) and PFS (C-index, 0.63 vs. 0.55; P = .04). Conclusion: CT texture analysis performed at baseline and early after starting PD-1/PD-L1 inhibitors is associated with clinical outcomes of patients with mRCC.
... TA was also performed without the filtration step on the conventional CT in the control group for comparison. This methodology has been previously described for technique and clinical application [12][13][14][15]. The current understanding of the pathophysiology is that micrometastases from the tumour that have entered into the hepatic circulation block portal flow. ...
Article
Full-text available
Background Radiomics allows information not readily available to the naked eye to be extracted from high resolution imaging modalities such as CT. Identifying that a cancer has already metastasised at the time of presentation through a radiomic signature will affect the treatment pathway. The ability to recognise the existence of metastases earlier will have a significant impact on the survival outcomes. Aim To create a novel radiomic signature using textural analysis in the evaluation of synchronous liver metastases in colorectal cancer. Methods CT images at baseline and subsequent surveillance over a 5-year period of patients with colorectal cancer were processed using textural analysis software. Comparison was made between those patients who developed liver metastases and those that remained disease free to detect differences in the ‘texture’ of the liver. Results A total of 24 patients were divided into two matched groups for comparison. Significant differences between the two groups scores when using the textural analysis programme were found on coarse filtration (p = 0.044). Patients that went on to develop metastases an average of 18 months after presentation had higher levels of hepatic heterogeneity on CT. Conclusion This initial study demonstrates the potential of using a textural analysis programme to build a radiomic signature to predict the development of hepatic metastases in rectal cancer patients otherwise thought to have clear staging CT scans at time of presentation.
... where P i, j is normalized symmetrical, N is number of grey levels, μ is mean intensity, σ is variance of intensity. The textural features were calculated using various equations used in (Castellano et al., 2004;Esgiar et al., 2002;Guru et al., 2010;Haider et al., 2017;Khalvati et al., 2015;Khuzi et al., 2008;Yu et al., 2017). The prediction size, radius of the prediction area, image equivalence, and dispersion were calculated using morphological features shown in Equations (6) and (7). ...
Article
Full-text available
Background The COVID-19 pandemic has precipitated global apprehensions about increased fatalities and raised concerns about gaps in healthcare infrastructure and accessibility the world over. Consequently, the importance of timely prediction and treatment of the disease to reduce transmission and mortality rates cannot be emphasized enough. Various symptoms of the disease have been identified as it progresses from the time it is contracted. COVID-19 has been found to internally affect the lungs, and the four progressive stages of the infection can be categorized as mild, moderate, severe, and critical. Therefore, an accurate analysis of the current stage of the disease that can help predict its progression has become critical. X-ray imaging has been found to be an effective screening procedure for predicting the various stages of this epidemic. Although many different approaches using machine learning, as well as deep learning were utilized to predict and classify diseases in general, till date, such an approach has not been used to predict the various stages of COVID-19 by using X-ray imaging to identify and classify those stages. Materials and method The proposed hybrid method used three public datasets for its implementation. In this work, extensive images were used for the purposes of testing and training. The dataset-1 consists of 1200 COVID-19 as well as 1200 Non-COVID-19 images, while dataset-2 used 700 COVID-19 as well as 700 Non-COVID-19 images, and finally, dataset-III utilized 1900 COVID-19 as well as 1900 Non-COVID-19 images for purposes of testing and training. The proposed work undertook the task of pre-processing using textual and morphological features, while the segmentation and prediction of COVID-19 as well as Non-COVID-19 images were undertaken using VGG-16 with light GBM for better prediction and handing of huge datasets, and finally, the classification of the various stages of COVID-19 images was performed using Deep Belief Network. Results The outcomes of the proposed work were subjected to several iterations which were then compared using different parameters such as accuracy, specificity, and sensitivity. In general, the prediction and grouping of the various stages of COVID-19 by using affected images were found to be 99.2%, 99.4% and 99.5%, respectively. The bacterial pneumonia prediction rates were observed to be 98.5%, 99.4% and 98.3%, respectively. The average classification of the stages were found to be 98.1%, 98.6% and 98.3%, while the combined multi-classification prediction rates were observed to be 98.6%, 99.1% and 98.7%, respectively.
... Using radiomic analysis, quantitative information can be extracted from medical images and analyzed to obtain a comprehensive understanding of the characteristics of individual tumors [18]. Recent studies have found that radiomic features have the potential to predict treatment response [19,20] and clinical outcomes [21][22][23][24]. ...
Article
Full-text available
Conventional methods to determine the response to immune checkpoint inhibitors (ICIs) are limited by the unique responses to an ICI. We performed a radiomics approach for all measurable lesions to identify radiomic variables that could distinguish hyperprogressive disease (HPD) on baseline CT scans and classify a dissociated response (DR). One hundred and ninety-six patients with advanced lung cancer, treated with ICI monotherapy, who underwent at least three CT scans, were retrospectively enrolled. For all 621 measurable lesions, HPDv was determined from baseline CT scans using the tumor growth kinetics (TGK) ratio, and radiomics features were extracted. Multivariable logistic regression analysis of radiomics features was performed to discriminate DR. Radiomics features that significantly discriminated HPDv on baseline CT differed according to organ. Of the 196 patients, 54 (27.6%) had a DR and 142 (72.4%) did not have a DR. Overall survival in the group with a DR was significantly inferior to that in the group without a DR (log rank test, p = 0.04). Our study shows that lesion-level analysis using radiomics features has great potential for discriminating HPDv and understanding heterogeneous tumor progression, including a DR, after ICI treatment.
Article
To achieve better treatment regimen and follow-up assessment design for intensity-modulated radiotherapy (IMRT) treated nasopharyngeal carcinoma (NPC) patients, an accurate progression free survival (PFS) time prediction algorithm is needed. We propose to develop PFS prediction model of NPC patients after IMRT treatment using deep learning method, and to compare that with traditional texture analysis method. 151 NPC patients were included in this retrospective study. T1 weighted, proton density and dynamic contrast enhanced MR images were acquired. Expression level of five genes (HIF-1α, EGFR, PTEN, Ki-67, VEGF) and infection of Epstein-Barr virus were tested. A residual network was trained to predict PFS from magnetic resonance (MR) images. The output as well as patient characteristics were combined using linear regression model to give a final PFS prediction. The prediction accuracy was compared with traditional texture analysis method. Regression model combining deep learning output with HIF-1α expression and EB infection gives the best PFS prediction accuracy (Spearman correlation R2 =0.53; Harrell's C-index = 0.82; ROC analysis AUC=0.88; log rank test HR= 8.45), higher than regression model combining texture analysis with HIF-1α expression (Spearman correlation R2 =0.14; Harrell's C-index =0.68; ROC analysis AUC=0.76; log rank test HR= 2.85). Deep learning method doesn't require manually drawn tumor ROI, thus is fully automatic. MR image processing using deep learning combined with patient characteristics can provide accurate PFS prediction for nasopharyngeal carcinoma patients and doesn't rely on specific kernels or tumor ROI as texture analysis method.
Article
Background: Whole-tumor apparent diffusion coefficient (ADC) histogram analysis is relevant to predicting the neoadjuvant chemoradiation therapy (nCRT) response in patients with locally advanced rectal cancer (LARC). Aim: To evaluate the performance of ADC histogram-derived parameters for predicting the outcomes of patients with LARC. Methods: This is a single-center, retrospective study, which included 48 patients with LARC. All patients underwent a pre-treatment magnetic resonance imaging (MRI) scan for primary tumor staging and a second restaging MRI for response evaluation. The sample was distributed as follows: 18 responder patients (R) and 30 non-responders (non-R). Eight parameters derived from the whole-lesion histogram analysis (ADCmean, skewness, kurtosis, and ADC10th, 25th, 50th, 75th, 90th percentiles), as well as the ADCmean from the hot spot region of interest (ROI), were calculated for each patient before and after treatment. Then all data were compared between R and non-R using the Mann-Whitney U test. Two measures of diagnostic accuracy were applied: the receiver operating characteristic curve and the diagnostic odds ratio (DOR). We also reported intra- and interobserver variability by calculating the intraclass correlation coefficient (ICC). Results: Post-nCRT kurtosis, as well as post-nCRT skewness, were significantly lower in R than in non-R (both P < 0.001, respectively). We also found that, after treatment, R had a larger loss of both kurtosis and skewness than non-R (∆%kurtosis and ∆skewness, P < 0.001). Other parameters that demonstrated changes between groups were post-nCRT ADC10th, ∆%ADC10th, ∆%ADCmean, and ROI ∆%ADCmean. However, the best diagnostic performance was achieved by ∆%kurtosis at a threshold of 11.85% (Area under the receiver operating characteristic curve [AUC] = 0.991, DOR = 376), followed by post-nCRT kurtosis = 0.78 × 10-3 mm2/s (AUC = 0.985, DOR = 375.3), ∆skewness = 0.16 (AUC = 0.885, DOR = 192.2) and post-nCRT skewness = 1.59 × 10-3 mm2/s (AUC = 0.815, DOR = 168.6). Finally, intraclass correlation coefficient analysis showed excellent intraobserver and interobserver agreement, ensuring the implementation of histogram analysis into routine clinical practice. Conclusion: Whole-tumor ADC histogram parameters, particularly kurtosis and skewness, are relevant biomarkers for predicting the nCRT response in LARC. Both parameters appear to be more reliable than ADCmean from one-slice ROI.
Chapter
Great advances have been achieved during the last decade in “Big Data” generation and the computational methods used to analyze these large datasets. The concepts and methods of Big Data and artificial intelligence were introduced into various medical specialties to improve management of patients. The high-throughput data collected from electronic health records, laboratory reports, radiological imaging, histopathological examination, gene expression profile, and any other patient-related sources are used to fuel the techniques of artificial intelligence in order to develop algorithms and models that are used for diagnosis, prediction, monitoring, or designing therapy. These advances are an active area of research, and the results seem promising, though practical application into routine clinical practice is still lagging behind. The use of Big Data in Urology has been investigated by several research groups, including addressing the diagnosis and treatment of urological cancers. The early diagnosis of renal cell cancer represents a challenge, and its incidence rate is rising in recent years, presumably due to increased incidental detection of renal masses during abdominal radiography for non-related complaints. In this chapter, we summarize the applications of Big Data generated from radiological imaging (radiomics), histopathological examination (pathomics), and molecular subtyping (genomics) and address how artificial intelligence was employed in the management of renal cell carcinoma.KeywordsGenomicsPathomicsRadiogenomicsRadiomicsRenal cell carcinoma
Article
Full-text available
Radiomics (radiogenomics) characterizes tumor phenotypes based on quantitative image features derived from routine radiologic imaging to improve cancer diagnosis, prognosis, prediction and response to therapy. Although radiomic features must be reproducible to qualify as biomarkers for clinical care, little is known about how routine imaging acquisition techniques/parameters affect reproducibility. To begin to fill this knowledge gap, we assessed the reproducibility of a comprehensive, commonly-used set of radiomic features using a unique, same-day repeat computed tomography data set from lung cancer patients. Each scan was reconstructed at 6 imaging settings, varying slice thicknesses (1.25 mm, 2.5 mm and 5 mm) and reconstruction algorithms (sharp, smooth). Reproducibility was assessed using the repeat scans reconstructed at identical imaging setting (6 settings in total). In separate analyses, we explored differences in radiomic features due to different imaging parameters by assessing the agreement of these radiomic features extracted from the repeat scans reconstructed at the same slice thickness but different algorithms (3 settings in total). Our data suggest that radiomic features are reproducible over a wide range of imaging settings. However, smooth and sharp reconstruction algorithms should not be used interchangeably. These findings will raise awareness of the importance of properly setting imaging acquisition parameters in radiomics/radiogenomics research.
Article
Full-text available
Magnetic Resonance Imaging (MRI) has been routinely used for the diagnosis and treatment of breast cancer. However, the relationship between the MRI tumor phenotypes and the underlying genetic mechanisms remains under-explored. We integrated multi-omics molecular data from The Cancer Genome Atlas (TCGA) with MRI data from The Cancer Imaging Archive (TCIA) for 91 breast invasive carcinomas. Quantitative MRI phenotypes of tumors (such as tumor size, shape, margin, and blood flow kinetics) were associated with their corresponding molecular profiles (including DNA mutation, miRNA expression, protein expression, pathway gene expression and copy number variation). We found that transcriptional activities of various genetic pathways were positively associated with tumor size, blurred tumor margin, and irregular tumor shape and that miRNA expressions were associated with the tumor size and enhancement texture, but not with other types of radiomic phenotypes. We provide all the association findings as a resource for the research community (available at http://compgenome.org/Radiogenomics/). These findings pave potential paths for the discovery of genetic mechanisms regulating specific tumor phenotypes and for improving MRI techniques as potential non-invasive approaches to probe the cancer molecular status.
Article
Full-text available
Analysis of texture within tumours on computed tomography (CT) is emerging as a potentially useful tool in assessing prognosis and treatment response for patients with cancer. This article illustrates the image and histological features that correlate with CT texture parameters obtained from tumours using the filtration-histogram approach, which comprises image filtration to highlight image features of a specified size followed by histogram analysis for quantification. Computer modelling can be used to generate texture parameters for a range of simple hypothetical images with specified image features. The model results are useful in explaining relationships between image features and texture parameters. The main image features that can be related to texture parameters are the number of objects highlighted by the filter, the brightness and/or contrast of highlighted objects relative to background attenuation, and the variability of brightness/contrast of highlighted objects. These relationships are also demonstrable by texture analysis of clinical CT images. The results of computer modelling may facilitate the interpretation of the reported associations between CT texture and histopathology in human tumours. The histogram parameters derived during the filtration-histogram method of CT texture analysis have specific relationships with a range of image features. Knowledge of these relationships can assist the understanding of results obtained from clinical CT texture analysis studies in oncology.
Article
Full-text available
Objective: Anti-angiogenic drugs cause a reduction in tumour density (Choi criteria) first and then in size [Response Evaluation Criteria In Solid Tumours (RECIST)]. The prognostic significance of changes in tumour density in metastatic renal cell carcinoma (mRCC) is unknown and was assessed in this study. Methods: The prognostic significance of partial response (PR) as opposed to non-response [stable disease (SD) + progressive (PD)] to anti-angiogenic therapy was assessed in patients with mRCC separately for both criteria using the log-rank test and Cox regression models. Results: Both criteria were applied to 35 patients. The response was identical for all eight patients with PR and most patients with PD (10/12) when using the RECIST and Choi criteria. Adding tumour density information, 14 patients with SD were re-categorised as having PR (7), SD (4), and PD (3). Patients with PR (Choi) were progression free significantly longer [hazard ratio (HR) 0.24; 95 % CI 0.10-0.57; P = 0.001] and had better overall survival (HR 0.36; 95 % CI 0.15-0.89; P = 0.026) compared to patients with SD or PD. The predictive value of PR according to RECIST was not statistically significant. Conclusions: In mRCC, the Choi criteria separate prognostic groups better when compared with RECIST. This may allow early discrimination of patients benefiting from continued treatment.
Article
Objective: The purpose of the present study is to determine whether CT texture features of newly diagnosed primary renal cell carcinomas (RCCs) correlate with pathologic features and oncologic outcomes. Materials and methods: CT texture analysis was performed on large (> 7 cm; mean size, 9.9 cm) untreated RCCs in 157 patients (52 women and 105 men; mean age, 60.3 years). Measures of tumor heterogeneity, including entropy, kurtosis, skewness, mean, mean of positive pixels, and SD of pixel distribution histogram were derived from multiphasic CT using various filter settings: unfiltered (spatial scaling factor, 0), fine (spatial scaling factor, 2), medium (spatial scaling factor, 3-4), or coarse (spatial scaling factor, 5-6). Texture values were correlated with histologic subtype, nuclear grade, pathologic stage, and clinical outcome. Results: When a coarse filter setting (spatial scaling factor, 6) was used, entropy on portal venous phase CT images was positively associated with clear cell histologic findings (odds ratio [OR], 134; 95% CI, 16-1110; p < 0.001) and was negatively associated with non-clear cell subtype findings (papillary spatial scale factor, 6; OR, 0.016; 95% CI, 0.002-0.132; p < 0.001). ROC curve analysis for entropy (on portal venous phase images obtained with a spatial scaling factor of 6) revealed an AUC of 0.943 (95% CI, 0.892-0.993) for clear cell histologic findings, with similar values noted for non-clear cell histologic findings. The mean of positive pixels and the SD of the pixel distribution histogram were statistically significantly associated with histologic cell type in a similar fashion. Entropy, the SD of the pixel distribution histogram, and the mean of positive pixels were associated with nuclear grade, most prominently when fine or medium texture filters were used (p < 0.05). There was a statistically significant association of texture features noted on unenhanced CT, including the SD of the pixel distribution histogram, the mean of positive pixels, and entropy, with the time to disease recurrence and death due to disease (e.g., for entropy noted on unenhanced CT images obtained with a spatial scaling factor of 6, the hazard ratio was 3.49 [95% CI, 1.55-7.84]; p = 0.002). Conclusion: CT texture features (in particular, entropy, the mean of positive pixels, and the SD of the pixel distribution histogram) are associated with tumor histologic findings, nuclear grade, and outcome measures. The contrast phase does seem to affect heterogeneity measures.
Article
The purpose of the study was to determine if CT texture features of untreated hepatic metastatic colorectal cancer (CRC) relate to pathologic features and clinical outcomes. Tumor texture analysis was performed on single hepatic metastatic lesions on pre-treatment contrast-enhanced CT scans in 77 pts (mean age 58, 34F/43M) using a novel tool. Measures of heterogeneity, including entropy, kurtosis, skewness, mean, mean positive pixels (MPP), and standard deviation (SD) of pixel distribution histogram were derived with filter values corresponding to fine (spatial scaling factor (ssf) 2), medium (ssf 3, 4), and coarse textures (ssf 5, 6). Texture parameters were correlated with tumor grade, baseline serum CEA, and KRAS mutation status. Overall survival was also correlated using Cox proportional hazards models. Single-slice 2D vs. whole-tumor volumetric 3D texture analysis was compared in a subcohort of 20 patients. Entropy, MPP, and SD at medium filtration levels were significantly associated with tumor grade (MPP ssf 3 P = 0.002, SD ssf 3 P = 0.004, entropy ssf 4 P = 0.007). Skewness was negatively associated KRAS mutation (P = 0.02). Entropy at coarse filtration levels was associated with survival (Hazard ratio (HR) for death 0.65, 95% CI 0.44-0.95, P = 0.03). Texture results for 2D and 3D analysis were similar. CT texture features, particularly entropy, MPP, and SD, are significantly associated with tumor grade in untreated CRC liver metastases. Tumor entropy at coarse filters correlates with overall survival. Single-slice 2D texture analysis appears to be adequate.
Article
Radiomics provides opportunities to quantify the tumor phenotype non-invasively by applying a large number of quantitative imaging features. This study evaluates computed-tomography (CT) radiomic features for their capability to predict distant metastasis (DM) for lung adenocarcinoma patients. We included two datasets: 98 patients for discovery and 84 for validation. The phenotype of the primary tumor was quantified on pre-treatment CT-scans using 635 radiomic features. Univariate and multivariate analysis was performed to evaluate radiomics performance using the concordance index (CI). Thirty-five radiomic features were found to be prognostic (CI>0.60, FDR<5%) for DM and twelve for survival. It is noteworthy that tumor volume was only moderately prognostic for DM (CI=0.55, p-value=2.77×10(-5)) in the discovery cohort. A radiomic-signature had strong power for predicting DM in the independent validation dataset (CI=0.61, p-value=1.79×10(-17)). Adding this radiomic-signature to a clinical model resulted in a significant improvement of predicting DM in the validation dataset (p-value=1.56×10(-11)). Although only basic metrics are routinely quantified, this study shows that radiomic features capturing detailed information of the tumor phenotype can be used as a prognostic biomarker for clinically-relevant factors such as DM. Moreover, the radiomic-signature provided additional information to clinical data. Copyright © 2015 Elsevier Ireland Ltd. All rights reserved.
Article
Purpose: To compare revised Choi criteria that incorporate concurrent size and attenuation changes at early follow-up imaging with Response Evaluation Criteria in Solid Tumors ( RECIST Response Evaluation Criteria in Solid Tumors ) 1.1 and original Choi criteria in stratification of clinical outcomes in patients with metastatic renal cell carcinoma ( mRCC metastatic renal cell carcinoma ) treated with sunitinib. Materials and methods: Institutional review board approved this retrospective study and waived informed consent. Baseline and first follow-up computed tomographic scans in 69 patients (50 men, 19 women; mean age, 60.3 years; range, 19-83 years) with mRCC metastatic renal cell carcinoma treated with sunitinib from October 1, 2008, to March 1, 2013, were evaluated for tumor response by using RECIST Response Evaluation Criteria in Solid Tumors 1.1, original Choi criteria, and revised Choi criteria. Correlations with overall survival ( OS overall survival ) and progression-free survival ( PFS progression-free survival ) were compared and stratified according to each radiologic criteria with Kaplan-Meier and multivariate Cox regression analysis. Results: Median follow-up time was 29.7 months (95% confidence interval [ CI confidence interval ]: 18.9, 45.9). Response according to revised Choi criteria was independently correlated with OS overall survival (hazard ratio, 0.47 [95% CI confidence interval : 0.23, 0.99]; P = .046) and PFS progression-free survival (hazard ratio, 0.53 [95% CI confidence interval : 0.29, 0.99]; P = .047). Response according to RECIST Response Evaluation Criteria in Solid Tumors was not significantly correlated with OS overall survival (hazard ratio, 0.65 [95% CI confidence interval : 0.27, 1.58]; P = .344) or PFS progression-free survival (hazard ratio, 0.89 [95% CI confidence interval : 0.42, 1.91]; P = .768). Response according to original Choi criteria was not significantly correlated with OS overall survival (hazard ratio, 0.60 [95% CI confidence interval : 0.32, 1.11]; P = .106) or PFS progression-free survival (hazard ratio, 0.59 [95% CI confidence interval : 0.34, 1.02]; P = .060). Median OS overall survival and PFS progression-free survival in responders according to revised Choi criteria was 39.4 months (95% CI confidence interval : 9.1, upper limit not estimated) and 13.7 months (95% CI confidence interval : 6.4, 24.6), respectively, compared with 12.8 months (95% CI confidence interval : 8.7, 18.0) and 5.3 months (95% CI confidence interval : 3.9, 8.4), respectively, in nonresponders. Conclusion: Contemporaneous reduction in tumor size and attenuation were correlated with favorable clinical outcomes. Response according to revised Choi criteria showed better correlation with clinical outcomes compared with that according to RECIST Response Evaluation Criteria in Solid Tumors or original Choi criteria in patients with mRCC metastatic renal cell carcinoma treated with sunitinib.
Article
Increased sunitinib exposure (area under the curve) is associated with better outcome in metastatic renal cell cancer. Recommendations for dose modification do not take this into account. A treatment strategy, based on individual patient toxicity, was developed to maximize dose and minimize time without therapy for patients who could not tolerate the standard sunitinib schedule of 50mg given for 28 days with a 14-day break (50mg, 28/14). A single-center retrospective review was conducted on patients with metastatic renal cell cancer treated from October 2005 to March 2010. Dose/schedule modifications (DSM) were done to keep toxicity (hematological, fatigue, skin, and gastrointestinal) at≤grade 2. DSM-1 was 50mg, 14 days on/7 days off with individualized increases in days on treatment. DSM-2 was 50mg, 7 days on/7 days off with individualized increase in days on treatment. DSM-3 was 37.5mg with individualized 7-day breaks. DSM-4 was 25mg with individualized 7-day breaks. Multivariable analysis was performed for outcome as a function of patient and treatment variables. Overall, 172 patients were included in the analysis. Most patients had clear cell histology (79.1%) with sunitinib given as a first-line therapy in 59%. The DSM-1 and 2 and DSM-3 and 4 groups had a progression-free survival (PFS) (10.9-11.9mo) and overall survival (OS) (23.4-24.5mo) that was significantly better than the PFS (5.3mo; P<0.001) and OS (14.4 mo; P = 0.03 and 0.003) for the standard schedule (50mg, 28/14). DCE-US in a subset of patients showed that maximum antiangiogenic activity was achieved after 14 days on therapy. Individualized sunitinib scheduling based on toxicity may improve PFS and OS. This hypothesis is supported by several other respective data that are reviewed. A confirmatory prospective trial is ongoing.
Article
Background: The International Metastatic Renal-Cell Carcinoma Database Consortium model offers prognostic information for patients with metastatic renal-cell carcinoma. We tested the accuracy of the model in an external population and compared it with other prognostic models. Methods: We included patients with metastatic renal-cell carcinoma who were treated with first-line VEGF-targeted treatment at 13 international cancer centres and who were registered in the Consortium's database but had not contributed to the initial development of the Consortium Database model. The primary endpoint was overall survival. We compared the Database Consortium model with the Cleveland Clinic Foundation (CCF) model, the International Kidney Cancer Working Group (IKCWG) model, the French model, and the Memorial Sloan-Kettering Cancer Center (MSKCC) model by concordance indices and other measures of model fit. Findings: Overall, 1028 patients were included in this study, of whom 849 had complete data to assess the Database Consortium model. Median overall survival was 18·8 months (95% 17·6-21·4). The predefined Database Consortium risk factors (anaemia, thrombocytosis, neutrophilia, hypercalcaemia, Karnofsky performance status <80%, and <1 year from diagnosis to treatment) were independent predictors of poor overall survival in the external validation set (hazard ratios ranged between 1·27 and 2·08, concordance index 0·71, 95% CI 0·68-0·73). When patients were segregated into three risk categories, median overall survival was 43·2 months (95% CI 31·4-50·1) in the favourable risk group (no risk factors; 157 patients), 22·5 months (18·7-25·1) in the intermediate risk group (one to two risk factors; 440 patients), and 7·8 months (6·5-9·7) in the poor risk group (three or more risk factors; 252 patients; p<0·0001; concordance index 0·664, 95% CI 0·639-0·689). 672 patients had complete data to test all five models. The concordance index of the CCF model was 0·662 (95% CI 0·636-0·687), of the French model 0·640 (0·614-0·665), of the IKCWG model 0·668 (0·645-0·692), and of the MSKCC model 0·657 (0·632-0·682). The reported versus predicted number of deaths at 2 years was most similar in the Database Consortium model compared with the other models. Interpretation: The Database Consortium model is now externally validated and can be applied to stratify patients by risk in clinical trials and to counsel patients about prognosis. Funding: None.