Periodic Reporting for period 1 - SKELETON-ID (Soft Computing and Computer Vision for Comparative Radiography in Forensic Identification)
Periodo di rendicontazione: 2018-04-01 al 2020-03-31
Forensic human identification is a great challenge in the preservation and defense of Human Rights. There is an urgent need to provide forensic practitioners with accurate, robust, unbiased and automatic identification systems. The MSCA IF Skeleton-ID fills part of this technological gap by automating the comparison of radiographs (CR) through a novel artificial intelligence paradigm for human identification.
CR traditionally involves the use of antemortem (AM) radiographs of the suspected deceased, producing postmortem (PM) radiographs that simulate the AM ones in scope and projection, and then performing a comparison looking for consistencies and inconsistencies in bone morphology, pathological and trauma conditions, etc. CR requires a prior record of clinical images that are not always available; but if present, this technique is extremely accurate, reaching >99% reliability for certain bones. Despite its proven validity for identification purposes, automatic and objective approaches are in their infancy in this field. Most existing proposals rely on the expert's skills and experience and follow an error-prone, time-consuming and subjective approach requiring PM X-ray acquisition in the same conditions of the AM one, and the manual delineation of the bone in AM and PM radiographies.
Skeleton-ID contributes to present the first complete automatic CR identification system, shifting from current observational methods to objective, fast, robust and reproducible ones. This system includes three stages: i) image segmentation: automatic delineation of the target bone's contour in the AM radiography; ii) image registration: automatic comparison of the PM 3D model of the bone and the delineated AM radiography; and iii) computer-aided decision support system able to integrate all available information and to assist the forensic expert in the decision making process.
More information regarding Skeleton-ID is available in the website of the project: https://www.ugr.es/~pmesejo/skeletonid.html
CR traditionally involves the use of antemortem (AM) radiographs of the suspected deceased, producing postmortem (PM) radiographs that simulate the AM ones in scope and projection, and then performing a comparison looking for consistencies and inconsistencies in bone morphology, pathological and trauma conditions, etc. CR requires a prior record of clinical images that are not always available; but if present, this technique is extremely accurate, reaching >99% reliability for certain bones. Despite its proven validity for identification purposes, automatic and objective approaches are in their infancy in this field. Most existing proposals rely on the expert's skills and experience and follow an error-prone, time-consuming and subjective approach requiring PM X-ray acquisition in the same conditions of the AM one, and the manual delineation of the bone in AM and PM radiographies.
Skeleton-ID contributes to present the first complete automatic CR identification system, shifting from current observational methods to objective, fast, robust and reproducible ones. This system includes three stages: i) image segmentation: automatic delineation of the target bone's contour in the AM radiography; ii) image registration: automatic comparison of the PM 3D model of the bone and the delineated AM radiography; and iii) computer-aided decision support system able to integrate all available information and to assist the forensic expert in the decision making process.
More information regarding Skeleton-ID is available in the website of the project: https://www.ugr.es/~pmesejo/skeletonid.html
The work performed during this Marie Curie Fellowship has been mainly related to the design, implementation and validation of machine learning and computer vision algorithms with forensic identification purposes. In particular, the final goal is to make progress towards an automatic comparative radiography (CR)-based decision support system to assist the forensic expert in the decision making process.
Within this general framework, three main research lines were carried out:
- We developed deep learning-based image segmentation algorithms to automatically delineate anatomical structures in X-ray images. The system developed was capable of providing human-competitive results in the segmentation of the heart, lungs, clavicles and frontal sinuses in radiographs.
- We designed a hierarchical fuzzy multi-evidence computer-aided decision support system (CADSS), which represents the first existing tool to assist the forensic anthropologist in the ID process using CR. This CADSS employs the image segmentation framework previously developed, together with image registration techniques, to compute the overlap between the antemortem (AM) and post-mortem (PM) materials. The results obtained showed that, with this preliminary system, we could rapidly reduce the number of potential candidates by 70%, keeping the remaining 30% to apply other complementary forensic identification approaches (therefore greatly reducing the time required for multiple comparisons).
- We extensively applied machine learning techniques to estimate the biological profile (mainly, sex and age) from bone remains. We also further investigated the methodological principles underlying the correct validation of prediction models.
The dissemination of the research carried out in this project includes scientific publications, participation in conferences, as well as educational activities for the general public (Week of Science, European Researchers' Night, and visits to schools and high-schools). All the results of this project will be exploited by Panacea Cooperative Research (https://panacea-coop.com/index.php/en/) as is the case of the two patent applications filed during the project. The same will happen with the advances made in the estimation of the biological profile by means of machine learning techniques. All these advances will be made available in the forensic identification framework developed by Panacea, termed Skeleton-ID (https://skeleton-id.com/index.php/en/) as add-ons or modules.
Within this general framework, three main research lines were carried out:
- We developed deep learning-based image segmentation algorithms to automatically delineate anatomical structures in X-ray images. The system developed was capable of providing human-competitive results in the segmentation of the heart, lungs, clavicles and frontal sinuses in radiographs.
- We designed a hierarchical fuzzy multi-evidence computer-aided decision support system (CADSS), which represents the first existing tool to assist the forensic anthropologist in the ID process using CR. This CADSS employs the image segmentation framework previously developed, together with image registration techniques, to compute the overlap between the antemortem (AM) and post-mortem (PM) materials. The results obtained showed that, with this preliminary system, we could rapidly reduce the number of potential candidates by 70%, keeping the remaining 30% to apply other complementary forensic identification approaches (therefore greatly reducing the time required for multiple comparisons).
- We extensively applied machine learning techniques to estimate the biological profile (mainly, sex and age) from bone remains. We also further investigated the methodological principles underlying the correct validation of prediction models.
The dissemination of the research carried out in this project includes scientific publications, participation in conferences, as well as educational activities for the general public (Week of Science, European Researchers' Night, and visits to schools and high-schools). All the results of this project will be exploited by Panacea Cooperative Research (https://panacea-coop.com/index.php/en/) as is the case of the two patent applications filed during the project. The same will happen with the advances made in the estimation of the biological profile by means of machine learning techniques. All these advances will be made available in the forensic identification framework developed by Panacea, termed Skeleton-ID (https://skeleton-id.com/index.php/en/) as add-ons or modules.
The increasing number of mass disasters, and the untraceable (and shameful) number of unidentified people worldwide, combined with new challenges due to mass migrations, pose human identification as a great challenge in the preservation and defense of human rights in today’s society. There is an urgent need to provide forensic practitioners with accurate, robust, unbiased and automate identification systems. The field of AI has the potential to develop those tools. However, there is an absence of industrial initiatives exploiting these emerging technologies that have so far curbed its application to practical identification scenarios.
The global forensic technologies market is expected to reach €88.9Bn by 2026 (9,6% CAGR2017-2026), thanks to increased government funding and rapid technological advancement of forensic technologies. The Skeleton-ID project is located in this dynamic and competitive environment, with the advantage that the solutions provided by Skeleton-ID and Panacea are the first of its kind and face no direct competition. Indirect competitors are DNA forensic solutions and AFIS (fingerprints comparison) providers. While these techniques are highly accurate, they are very time-consuming (several days/weeks in the case of DNA analysis) or expensive (in the order of millions of euros, in the case of AFIS). Also, they have a limited range of applicability (i.e. need DNA databases or soft tissue preservation). Therefore, the solutions provided by this Marie Curie project represent a cheaper and faster alternative, applicable when other techniques are not (e.g. decomposed bodies, lack of trustable records).
Personal identity is associated with the preservation and defense of Human Rights (HR) and is a tool to repair the violation of these rights. The Universal Declaration of HR was proclaimed by the United Nations General Assembly in Paris on December 10th 1948. Since then, a great number of actions towards the investigation of crimes against humanity have been carried out all over the world, in both, conflict and post-conflict environments. One essential step in this direction is the identification of missing persons. In this sense, the European Union has a strategic framework on HR. The Charter of Fundamental Rights of the European Union (adopted in 2000 and binding on EU countries since 2009) complements national HR documents and the European Convention on Human Rights (ECHR). Its purpose is to protect fundamental rights within the EU and to promote them worldwide under the values of Human dignity, freedom, democracy, equality, the rule of law and respect for HR. Nevertheless, as the authors of the 2017 Fundamental Rights Report in Europe indicate in the 10th anniversary of the European Union Agency for Fundamental Rights, “profound gaps in the implementation of fundamental rights persist on the ground and - in some areas - are deepening”.
The global forensic technologies market is expected to reach €88.9Bn by 2026 (9,6% CAGR2017-2026), thanks to increased government funding and rapid technological advancement of forensic technologies. The Skeleton-ID project is located in this dynamic and competitive environment, with the advantage that the solutions provided by Skeleton-ID and Panacea are the first of its kind and face no direct competition. Indirect competitors are DNA forensic solutions and AFIS (fingerprints comparison) providers. While these techniques are highly accurate, they are very time-consuming (several days/weeks in the case of DNA analysis) or expensive (in the order of millions of euros, in the case of AFIS). Also, they have a limited range of applicability (i.e. need DNA databases or soft tissue preservation). Therefore, the solutions provided by this Marie Curie project represent a cheaper and faster alternative, applicable when other techniques are not (e.g. decomposed bodies, lack of trustable records).
Personal identity is associated with the preservation and defense of Human Rights (HR) and is a tool to repair the violation of these rights. The Universal Declaration of HR was proclaimed by the United Nations General Assembly in Paris on December 10th 1948. Since then, a great number of actions towards the investigation of crimes against humanity have been carried out all over the world, in both, conflict and post-conflict environments. One essential step in this direction is the identification of missing persons. In this sense, the European Union has a strategic framework on HR. The Charter of Fundamental Rights of the European Union (adopted in 2000 and binding on EU countries since 2009) complements national HR documents and the European Convention on Human Rights (ECHR). Its purpose is to protect fundamental rights within the EU and to promote them worldwide under the values of Human dignity, freedom, democracy, equality, the rule of law and respect for HR. Nevertheless, as the authors of the 2017 Fundamental Rights Report in Europe indicate in the 10th anniversary of the European Union Agency for Fundamental Rights, “profound gaps in the implementation of fundamental rights persist on the ground and - in some areas - are deepening”.