ArticlePublisher preview available

Identification score for robust and secure identification using ante- and post-mortem skull CT scans

Authors:
Marie-Edith Richard
Marie-Edith Richard
Marie-Edith Richard
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Corentin Provost at Centre Hospitalier Sainte Anne
Corentin Provost
Tania Delabarde at Institut Médico-légal Paris
Tania Delabarde
  • Institut Médico-légal Paris
Pauline Iorio
Pauline Iorio
Pauline Iorio
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Show all 9 authorsHide
Read publisher preview
Download citation
To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

Due to their unique anatomy, paranasal sinuses have been used for comparative identification between post-mortem CT (PMCT) and ante-mortem CT (AMCT). However, data security issues arise when transferring raw AMCT images of a suspected identity. The aim of this study was to derive and validate an identification score based on CT slices extracted from successive CTs for the identification of subjects. For derivation procedure, we included patients who underwent two successive AMCTs at ≥ 1-year interval (n = 98), and 4 radiologists individually assessed similarity of prespecified CT slices (centered on ethmoid, frontal sinus and Left Semi-Circular Canal). Predictive values were calculated for all combinations of number of readers and slices, and the optimal compromise, termed IDScore, was selected. For validation, we included PMCTs performed between 2018 and 2022 with available comparative head AMCTs (n = 27). For each PMCT, 5 comparison procedures were performed: 1 concordant (with corresponding AMCT) and 4 discordant (with randomly selected AMCTs). Two radiologists evaluated similarity of ethmoid and frontal CT slices with a score ranging from -2 to + 2. IDScore was defined as the sum of these slice scores, averaged between the two readers. In the 135 comparison procedures, IDScore using predetermined thresholds (positive identification for IDScore > + 2, negative identification for IDScore < -1) allowed a perfect discrimination between identical subjects (Sensitivity = 100%, Specificity = 100%). IDScore could be used for remote identification of a subject with no need to access to the complete raw AMCT images, hence helping to overcome ethical and regulatory issues to access AMCT of a suspected identity. Trial registration: F20220729161623 on Health Data Hub, registered on 29 July 2022.
Procedure for slices positioning. All slices were positioned using double obliquity reformatting, using as reference the ethmoid roof on sagittal slice (a), and the largest width of frontal sinus on coronal slice (b). The Ethmoid slice (d) was captured 1–2 mm below the ethmoid roof (white line in a), and the Frontal slice (e) at the location where the frontal sinus was the largest. The LSCC slice (f) was captured on a plane tangent to both LSCCs (white line in c). The obtained slices d, e and f were used in the study
Procedure for slices positioning. All slices were positioned using double obliquity reformatting, using as reference the ethmoid roof on sagittal slice (a), and the largest width of frontal sinus on coronal slice (b). The Ethmoid slice (d) was captured 1–2 mm below the ethmoid roof (white line in a), and the Frontal slice (e) at the location where the frontal sinus was the largest. The LSCC slice (f) was captured on a plane tangent to both LSCCs (white line in c). The obtained slices d, e and f were used in the study
… 
Predictive values of CScores according to number of planes and readers. For each CScore, defined by a combination of the number of planes evaluated and the readers involved, we computed the negative and positive predictive values for assuming positive identification across multiple procedures. Each procedure corresponded to a different combination of plane types (frontal, ethmoid, and/or LSCC) and readers (1, 2, 3, and/or 4). From these computations, only the minimal predictive values across all procedures were retained to generate this graph
Predictive values of CScores according to number of planes and readers. For each CScore, defined by a combination of the number of planes evaluated and the readers involved, we computed the negative and positive predictive values for assuming positive identification across multiple procedures. Each procedure corresponded to a different combination of plane types (frontal, ethmoid, and/or LSCC) and readers (1, 2, 3, and/or 4). From these computations, only the minimal predictive values across all procedures were retained to generate this graph
… 
Example of negative identification with anatomical similarity between frontal sinuses. Frontal slices (a, b) and Ethmoid slices (c, d) of an AMCT (a, c) and a PMCT (b, d) are presented. Note the anatomical proximity between Frontal features (arrow in b). Obvious differences between ethmoid slices (overall shape, large left posterior cell on c) were used for negative identification. IDScore1 was -4, IDScore2 was -2 and IDScore was -3
Example of negative identification with anatomical similarity between frontal sinuses. Frontal slices (a, b) and Ethmoid slices (c, d) of an AMCT (a, c) and a PMCT (b, d) are presented. Note the anatomical proximity between Frontal features (arrow in b). Obvious differences between ethmoid slices (overall shape, large left posterior cell on c) were used for negative identification. IDScore1 was -4, IDScore2 was -2 and IDScore was -3
… 
Example of positive identification with slice positioning variation. Frontal slices (a, b) and Ethmoid slices (c, d) of an AMCT (a, c) and a PMCT (b, d) are presented. Note the perfect concordance between Frontal slices for septa position, but not for septa length. The posterior ethmoid cells seem much larger on c (white arrow) as compared to d due to registration error. However, most ethmoidal features are easily recognized in the anterior ethmoid (arrow in d). IDScore1 was 3, IDScore2 was 4 and IDScore was 3.5. The slice positioning was not perfect in this example, however this underlines that the IDScore should not be used as a definitive identification method, but rather a temporary identification method in a context of high-probability identity check, enabling data transfer and definitive identification by standard volume coregistration
Example of positive identification with slice positioning variation. Frontal slices (a, b) and Ethmoid slices (c, d) of an AMCT (a, c) and a PMCT (b, d) are presented. Note the perfect concordance between Frontal slices for septa position, but not for septa length. The posterior ethmoid cells seem much larger on c (white arrow) as compared to d due to registration error. However, most ethmoidal features are easily recognized in the anterior ethmoid (arrow in d). IDScore1 was 3, IDScore2 was 4 and IDScore was 3.5. The slice positioning was not perfect in this example, however this underlines that the IDScore should not be used as a definitive identification method, but rather a temporary identification method in a context of high-probability identity check, enabling data transfer and definitive identification by standard volume coregistration
… 
Proposed workflow for assessing IDScore with two radiologists Anonymized Ethmoid and Frontal slices are transferred between two radiologists for computing IDScore1 and IDScore2. The IDScore result is then used for lifting anonymity if > 2. A secure online meeting between radiologists for slice-to-slice comparison is proposed in the cases where IDScore result is uncertain
Proposed workflow for assessing IDScore with two radiologists Anonymized Ethmoid and Frontal slices are transferred between two radiologists for computing IDScore1 and IDScore2. The IDScore result is then used for lifting anonymity if > 2. A secure online meeting between radiologists for slice-to-slice comparison is proposed in the cases where IDScore result is uncertain
… 
Figures - available from: International Journal of Legal Medicine
This content is subject to copyright. Terms and conditions apply.
A preview of this full-text is provided by Springer Nature.
Learn more
Content available from International Journal of Legal Medicine
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
International Journal of Legal Medicine (2025) 139:639–649
https://doi.org/10.1007/s00414-024-03361-6
ORIGINAL ARTICLE
Identification score forrobust andsecure identification using ante‑
andpost‑mortem skull CT scans
Marie‑EdithRichard1,2· CorentinProvost1,2· TaniaDelabarde3,4,5· PaulineIorio6,7· YvesMenu6,7· GhaziHmeydia1,2·
BertrandLudes3,4,5· CatherineOppenheim1,2· JosephBenzakoun1,2
Received: 15 February 2024 / Accepted: 23 October 2024 / Published online: 4 November 2024
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024
Abstract
Due to their unique anatomy, paranasal sinuses have been used for comparative identification between post-mortem CT
(PMCT) and ante-mortem CT (AMCT). However, data security issues arise when transferring raw AMCT images of a
suspected identity. The aim of this study was to derive and validate an identification score based on CT slices extracted
from successive CTs for the identification of subjects. For derivation procedure, we included patients who underwent two
successive AMCTs at ≥ 1-year interval (n = 98), and 4 radiologists individually assessed similarity of prespecified CT slices
(centered on ethmoid, frontal sinus and Left Semi-Circular Canal). Predictive values were calculated for all combinations
of number of readers and slices, and the optimal compromise, termed IDScore, was selected. For validation, we included
PMCTs performed between 2018 and 2022 with available comparative head AMCTs (n = 27). For each PMCT, 5 comparison
procedures were performed: 1 concordant (with corresponding AMCT) and 4 discordant (with randomly selected AMCTs).
Two radiologists evaluated similarity of ethmoid and frontal CT slices with a score ranging from -2 to + 2. IDScore was
defined as the sum of these slice scores, averaged between the two readers. In the 135 comparison procedures, IDScore
using predetermined thresholds (positive identification for IDScore > + 2, negative identification for IDScore < -1) allowed a
perfect discrimination between identical subjects (Sensitivity = 100%, Specificity = 100%). IDScore could be used for remote
identification of a subject with no need to access to the complete raw AMCT images, hence helping to overcome ethical and
regulatory issues to access AMCT of a suspected identity.
Trial registration: F20220729161623 on Health Data Hub, registered on 29 July 2022.
Keywords Computed tomography· Human identification· Sinus· Forensic radiology
Introduction
Identification is a major issue in forensic context. Inter-
national recommendations have been made by Interpol to
positively match identification, in particular in disaster vic-
tim identification [1]. The criteria for positive identifica-
tion are based on primary methods consisting of friction
ridge analysis, odontology and DNA analysis and secondary
methods mostly based on personal data and medical findings
[2]. Forensic imaging, especially computed tomography is
currently increasingly used [3]. It provides information on
death causes [4, 5] but can also be used for identification,
to get biological characteristics such as age, sex and stature
[6, 7]. It also provides information on medical status, such
as surgery history with orthopedic implants, fractures or
metastases [6].
* Joseph Benzakoun
j.benzakoun@ghu-paris.fr
1 Institute ofPsychiatry andNeuroscience ofParis (IPNP),
Université Paris Cité, INSERM U1266, 75014Paris, France
2 Service de Radiologie, GHU Paris Psychiatrie Et
Neurosciences, Site Sainte-Anne, 1, Rue Cabanis,
75014Paris, France
3 Institut Médico-Légal de Paris, Paris, France
4 Université Paris Cité, UMR8045 BABEL, CNRS, Paris,
France
5 Pôle Universitaire d’imagerie Post-Mortem, Université Paris
Cité, Paris, France
6 Department ofRadiology, Hôpital Saint-Antoine, APHP,
Paris, France
7 Paris Sorbonne Université, Paris, France
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
ResearchGate has not been able to resolve any citations for this publication.
Human identification performed with skull’s sphenoid sinus based on deep learning
Article
Full-text available
  • Jul 2022
  • INT J LEGAL MED
Human identification plays a significant role in the investigations of disasters and criminal cases. Human identification could be achieved quickly and efficiently via 3D sphenoid sinus models by customized convolutional neural networks. In this retrospective study, a deep learning neural network was proposed to achieve human identification of 1475 noncontrast thin-slice CT scans. A total of 732 patients were retrieved and studied (82% for model training and 18% for testing). By establishing an individual recognition framework, the anonymous sphenoid sinus model was matched and cross-tested, and the performance of the framework also was evaluated on the test set using the recognition rate, ROC curve and identification speed. Finally, manual matching was performed based on the framework results in the test set. Out of a total of 732 subjects (mean age 46.45 years ± 14.92 (SD); 349 women), 600 subjects were trained, and 132 subjects were tested. The present automatic human identification has achieved Rank 1 and Rank 5 accuracy values of 93.94% and 99.24%, respectively, in the test set. In addition, all the identifications were completed within 55 s, which manifested the inference speed of the test set. We used the comparison results of the MVSS-Net to exclude sphenoid sinus models with low similarity and carried out traditional visual comparisons of the CT anatomical aspects of the sphenoid sinus of 132 individuals with an accuracy of 100%. The customized deep learning framework achieves reliable and fast human identification based on a 3D sphenoid sinus and can assist forensic radiologists in human identification accuracy.
View
Multisite Comparison of MRI Defacing Software Across Multiple Cohorts
Article
Full-text available
  • Feb 2021
With improvements to both scan quality and facial recognition software, there is an increased risk of participants being identified by a 3D render of their structural neuroimaging scans, even when all other personal information has been removed. To prevent this, facial features should be removed before data are shared or openly released, but while there are several publicly available software algorithms to do this, there has been no comprehensive review of their accuracy within the general population. To address this, we tested multiple algorithms on 300 scans from three neuroscience research projects, funded in part by the Ontario Brain Institute, to cover a wide range of ages (3–85 years) and multiple patient cohorts. While skull stripping is more thorough at removing identifiable features, we focused mainly on defacing software, as skull stripping also removes potentially useful information, which may be required for future analyses. We tested six publicly available algorithms (afni_refacer, deepdefacer, mri_deface, mridefacer, pydeface, quickshear), with one skull stripper (FreeSurfer) included for comparison. Accuracy was measured through a pass/fail system with two criteria; one, that all facial features had been removed and two, that no brain tissue was removed in the process. A subset of defaced scans were also run through several preprocessing pipelines to ensure that none of the algorithms would alter the resulting outputs. We found that the success rates varied strongly between defacers, with afni_refacer (89%) and pydeface (83%) having the highest rates, overall. In both cases, the primary source of failure came from a single dataset that the defacer appeared to struggle with - the youngest cohort (3–20 years) for afni_refacer and the oldest (44–85 years) for pydeface, demonstrating that defacer performance not only depends on the data provided, but that this effect varies between algorithms. While there were some very minor differences between the preprocessing results for defaced and original scans, none of these were significant and were within the range of variation between using different NIfTI converters, or using raw DICOM files.
View
State of the art in post-mortem computed tomography: a review of current literature
Article
Full-text available
  • Aug 2019
  • Virchows Arch
Computed tomography (CT) and other advanced diagnostic imaging techniques are gaining popularity in forensic pathology. This paper aims to define and offer complete and easily accessible “state of the art” for post-mortem computed tomography (PMCT), by reviewing the latest international literature. The proposed format answers the “five Ws” that follows: (1) What: We report the different kinds of CT scan and settings generally used in post-mortem imaging. The machine most employed is a 8/16-slice spiral CT, usually without contrast enhancement. The introduction of some variables, such as CT-guided biopsies, post-mortem ventilation, and PMCT angiography is becoming increasingly useful. (2) Why: Literature highlights the many advantages of PMCT. Limitations can be partly overcome by modern imaging techniques and combined evaluation with traditional autopsy. (3) Who: Most authors agree that collaboration between different specialists, i.e., radiologists and pathologists, is the best scenario, since radiologic, anatomic, and forensic skills are needed simultaneously. The most important human factor is “teamwork”. (4) When: Literature provides no absolute limits for performing PMCT. Some authors have tested PMCT as a replacement for conventional autopsy but found some limitations. Others evaluated PMCT as a guide or screening tool for traditional autopsy. (5) Where: Many research groups around the world have performed studies on the use of PMCT. Although few countries adopt PMCT in routine practice, its use is rapidly spreading.
View
The role of forensic anthropology in disaster victim identification (DVI): recent developments and future prospects
Article
Full-text available
  • Oct 2018
Forensic anthropological knowledge has been used in disaster victim identification (DVI) for over a century, but over the past decades, there have been a number of disaster events which have seen an increasing role for the forensic anthropologist. The experiences gained from some of the latest DVI operations have provided valuable lessons that have had an effect on the role and perceived value of the forensic anthropologist as part of the team managing the DVI process. This paper provides an overview of the ways in which forensic anthropologists may contribute to DVI with emphasis on how recent experiences and developments in forensic anthropology have augmented these contributions. Consequently, this paper reviews the value of forensic anthropological expertise at the disaster scene and in the mortuary, and discusses the way in which forensic anthropologists may use imaging in DVI efforts. Tissue-sampling strategies for DNA analysis, especially in the case of disasters with a large amount of fragmented remains, are also discussed. Additionally, consideration is given to the identification of survivors; the statistical basis of identification; the challenges related to some specific disaster scenarios; and education and training. Although forensic anthropologists can play a valuable role in different phases of a DVI operation, they never practice in isolation. The DVI process requires a multidisciplinary approach and, therefore, has a close collaboration with a range of forensic specialists.
View
An innovative 3D-3D superimposition for assessing anatomical uniqueness of frontal sinuses through segmentation on CT scans
Article
Full-text available
  • Jul 2019
  • INT J LEGAL MED
Anatomical uniqueness plays a significant role in the personal identification process of unknown deceased. Frontal sinuses have been widely used in the past decades for this purpose, mostly using 2D X-ray techniques. However, the modern 3D CT-based segmentation methods may help in developing novel and more reliable methods of identification. This study aims at assessing the anatomical uniqueness of frontal sinuses through the 3D model registration. Thirty subjects who underwent two maxillofacial CT scans (interval: 1 month to 5 years) were selected from a hospital database. Frontal sinuses were automatically segmented through ITK-SNAP open source software and the 3D models belonging to the same patient were automatically superimposed according to the least point-to-point difference between the two surfaces. Two hundred patients were randomly selected from the same database and undergo the same procedure to perform 200 superimpositions of frontal sinuses belonging to different individuals, equally divided between males and females (mismatches). Statistically significant differences of average root mean square (RMS) point-to-point distance between the group of matches and mismatches, as well as possible differences according to sex, were assessed through Mann-Whitney U test (p < 0.05). In the group of matches, RMS ranged between 0.07 and 0.96 mm (mean RMS 0.35 ± 0.23 mm), while in the group of mismatches, it ranged between 0.96 and 10.29 mm (mean RMS 2.59 ± 1.79 mm), with a statistically significant difference (p < 0.0001). Neither the matches nor the mismatches group showed statistically significant differences according to sex. This study proposes a novel 3D approach for the assessment of anatomical uniqueness of frontal sinuses, providing both morphological and quantitative analysis, and a new method of identification based on 3D assessment of frontal sinuses, applicable when ante-mortem CT scans are available.
View
Forensic Identification from Three-Dimensional Sphenoid Sinus Images Using the Iterative Closest Point Algorithm
Article
  • Apr 2022
Forensic identification of human remains is crucial for legal, humanitarian, and civil reasons. Wide heterogeneity in sphenoid sinus morphology can be used for personal identification. This study aimed to propose a new protocol for personal identification based on three-dimensional (3D) reconstruction of sphenoid sinus CT images using Iterative Closest Point (ICP) algorithm. Seven hundred thirty-two patients which consisted of 348 females and 384 males were retrospectively included. The study sample includes 732 previous images as a source point set and 743 later ones as a scene target set. The sphenoid sinus computed tomography (CT) images were processed on a workstation (Dolphin imaging) to obtain 3D images and stored as a file format of Stereo lithography (.STL). Then, a Python library vtkplotter was used to transform the STL format to PLY format, which was adapted to Point Cloud Library (PCL). The ICP algorithm was used for point clouds matching. The metric Rank-N recognition rate was used for evaluation. The scene target set of 743 individuals was compared with the source point set of 732 individual models and achieved Rank-1 accuracy of 96.24%, Rank-2 accuracy of 99.73%, and Rank-3 accuracy of 100%. Our results indicated that the 3D point cloud registration of sphenoid sinuses was useful for assessing personal identification in forensic contexts.
View
Human identification: an investigation of 3D models of paranasal sinuses to establish a biological profile on a modern UK population
Conference Paper
  • Jun 2021
Forensic anthropology traditionally aims to assist law enforcement with human identification by physically examining skeletal remains and assigning a biological profile using various metric and visual methods. These methods are crucial when a body undergoes extreme damage and standard approaches for positive identification are not possible. However, the traditional methods employed by forensic anthropologists were primarily developed from North American reference populations and have demonstrated varying accuracy rates when assigning age, sex, and ancestry to individuals outside of the reference collection. Medical imaging is a valuable source for facilitating empirical research and an accessible gateway for developing novel forensic anthropological methods for analysis including 3D modelling. This is especially critical for the United Kingdom (UK) where biological profiling methods developed from modern UK populations do not currently exist. Researchers have quantified the variability of the paranasal sinuses between individuals and have begun to explore their ability to provide biological information. However, the published literature that addresses these structures in a forensic context presents extremely varied insights and to date there has been no standardisation. This thesis presents research that addresses this gap and introduces a new approach for human identification using 3D models of the paranasal sinuses. The models were produced from a database of modern CT scans provided by University College London Hospital (UCLH), London, UK. Linear measurements and elliptic Fourier coefficients taken from 1,500 three-dimensional models across six ethnic groups assessed by one-way ANOVA and discriminant function analysis showed a range of classification rates with certain rates reaching 75-85.7% (p<0.05) in correctly classifying age and sex according to size and shape. The findings offer insights into the potential for employing CT scans to develop identification methods within the UK and establishes a foundation for using the paranasal sinuses as an attribute for establishing identification of unknown human remains in future crime reconstructions.
View
Frontal sinuses as tools for human identification: a systematic review of imaging methods
Article
  • Apr 2021
  • DENTOMAXILLOFAC RAD
The frontal sinuses are potential evidences for human identification because of the inherent distinctiveness of their morphology. Over the last decades, several techniques emerged to enable the visualization and analysis of the frontal sinuses via bi- and three-dimensional imaging. This systematic review aimed to compile different methodological approaches found in the scientific literature to contribute to human identification. Three examiners revisited the scientific literature in order to find imaging techniques for the visualization of the frontal sinuses applied to human identification. The standard search strings built-up from a PICO question identified 404 unique articles in the following databases Medline/Pubmed, Web of Science, Scopus, Lilacs and Scielo. Based on eligibility criteria applied during title, abstract and full-text reading, the sample reduced to 19 articles. The articles were published between 1987 and 2019 by research groups from 10 different countries. Computed tomography was used in 37% of the techniques, while the remaining (63%) techniques used skull radiographs. The techniques were highly heterogeneous and varied between metric analysis, direct image superimposition and morphology code-based systems. The authors considered their techniques useful for human identification and reported accuracy rates from 13 to 100%. Most of the studies revealed low risk of bias. More advantages were related with the techniques based on direct image superimpositions and three-dimensional visualization. Forensic experts must be aware of the use of frontal sinuses for human identification, especially when three-dimensional images are available as ante-mortem and post-mortem evidences for superimposition and comparison.
View
Post-mortem X-ray computed tomography (PMCT) identification using ante-mortem CT-scan of the sphenoid sinus
Article
  • Jul 2019
  • J NEURORADIOLOGY
Abstract PURPOSE: To evaluate forensic identification of individuals through visual comparison of sphenoid sinus anatomical configuration using ante- and post-mortem CT-scans. METHOD AND MATERIALS: Ante and post mortem head CT-scan of 33 individuals were retrospectively collected. Ten head CT-scans were randomly selected from various neurological contexts and added to the ante mortem group. Ten other head CT-scans were randomly selected from our post mortem PACS and added to the post mortem group. These CT-scans were assigned into 2 groups for analysis: an ante mortem group (33 + 10) and a post mortem group (33 + 10). For ethics and to avoid identification bias, CT-scans were anonymized – not showing any head structure but only sphenoid sinuses. An anatomical based classification system using the sphenoid sinuses anatomical variations was created according to anatomical and surgical literature. This classification was used by readers to identify in two different steps a maximum of matched and then unmatched scans. RESULTS: The first reader had a sensitivity of 100%[CI: 89.4%-100%] and a specificity of 100%[CI:99.8%-100%]. Sensitivity and specificity were respectively 93.9%[CI:79.8%-99.3%] and 99.9%[CI:99.6%-100%] for the second reader. Positive and negative predictive values were respectively 100%[CI: 89.4%-100%] and 100%[CI:99.8%-100%] for the first reader. Positive and negative values were respectively 96.9%[CI:83.8%-99.9%] and 99.9%[CI:99.7%-100%] for the second reader. Inter reader variability was estimated by Cohen’s kappa and an excellent agreement was found. CONCLUSION: We reported an excellent validity and reliability of subjective visual comparison of ante- and post mortem CT-data using an anatomical based classification of the sphenoid sinus. Keywords sphenoid sinusforensic identificationpost mortem computed tomography
View
Application of 3D image fusion for radiological identification of decedents
Article
  • Apr 2018
The fusion of antemortem and postmortem datasets may facilitate the radiological identification of decedents (RadID). Software for the image fusion is readily available from clinical radiology. This article demonstrates image fusions of three-dimensional datasets using medical image fusion software and describes RadID based on paranasal sinus, orthopedic implant, degenerative changes, healed fracture and dental positions by means of superimposed datasets (from computed tomography and magnetic resonance imaging).
View