FAME: Fully Automatic MRI-based Age Estimation of Adolescents

The aim of the proposed project FAME is research on novel methods for radiological age estimation of adolescents. Both, in a clinical context as well as in forensic applications, here especially in age estimation of minor unaccompanied adolescent asylum seekers without valid identification documents, a great need for a modern radiological age estimation method exists. This method is supposed to be free of harmful ionizing radiation, which is not the case for the established methods proposed in literature making use of X-ray and Computed Tomography. In forensic applications involving healthy subjects, the use of ionizing radiation is legally prohibited, thus research on an alternative method is highly relevant. Especially in age estimation of unaccompanied asylum seekers, which is a problem area of growing interest due to increased numbers of displaced people seeking shelter in European countries, these considerations are crucial. Magnetic Resonance Imaging promises an alternative for radiological age estimation. Besides the imaging technique being more applicable, since it is free of harmful ionizing radiation, it also enables improved accuracy, due to the depiction of 3D information and its better ability to visualize soft tissue structures as opposed to 2D X-ray projections. Together with modern medical computer vision methods based on machine learning techniques, an improvement in objectivity and repeatability is to be expected as well, since automatic algorithms for age estimation pose benefits compared to the subjective comparison of data to qualitative, sometimes outdated radiological atlases and staging schemes. Research goal of the proposed inter-disciplinary project FAME is to establish an automated, software-based multi-factorial age estimation method based on MRI data of the hand, the clavicle, and the third molars. The project will combine the competencies from the Ludwig Boltzmann Institute for Clinical Forensic Imaging in Graz, where forensic age assessments based on established X-ray based imaging modalities are routinely performed for the Austrian government, and the Institute for Computer Graphics and Vision at Graz University of Technology, where expertise on machine learning and medical computer vision is available. This novel, automated method, that incorporates contributions in landmark localization, age regression from MRI data, interaction between radiologists and algorithms for feature selection, and multi-estimator fusion, is intended to replace the established age estimation procedure that determines the age of unaccompanied minor asylum seekers who lack valid identification documents. It resembles a crucial first research step on a methodological alternative to the X-ray based methods and, after investigation in a clinical validation study on a larger cohort of subjects with potentially varying ethnicity, is building the foundation for the long-term goal of establishing a new gold standard for forensic as well as clinical age estimation.

In the course of the FWF project FAME Fully Automatic MRI-based Age Estimation of Adolescents, the Ludwig Boltzmann Institute for Clinical Forensic Imaging in Graz and the Institute for Computer Graphics and Vision at Graz University of Technology have developed a software that allows the automatic estimation of chronological age from magnetic resonance images (MRI) of adolescents. Chronological age estimation of adolescents has seen a lot of interest recently, due to migration movements into Europe. One of the challenges in these migration movements is to determine if asylum seekers, who do not possess valid identification documents, legally are minors or grown-ups. The state-of-the-art method in Austria, but also in other European countries, is the assessment of the skeletal and dental development through X-ray and computed tomography (CT) imaging. This allows an estimation of a very likely minimum age of a person. In the FWF project FAME, we have improved upon this technique by (1) using MRI that does not involve any potentially harmful ionizing radiation being used for healthy subjects, (2) three-dimensionally analyzing anatomical structures that show developmental progress (hand, clavicle, wisdom teeth) during aging, and (3) performing an automated, software-based age estimation that is objective and reproducible. The developed age estimation software required basic research in medical image analysis. In the course of the project we have developed novel methods for automatically localizing anatomical structures from MRI data. Furthermore, we were the first group to propose a prediction software for multi-factorial (hand, clavicle, wisdom teeth) chronological age estimation in the research literature. Our novel methods make extensive use of current machine learning techniques (deep learning) to solve these medical image analysis problems. To achieve that, we have used an MRI dataset of 320 volunteering Caucasian subjects, which was collected at the Ludwig Boltzmann Institute for Clinical Forensic Imaging, to train these machine learning algorithms. Our developed methods now enable the validation of a larger, more heterogeneous dataset, to be used in practice in the asylum seeking procedure. This requires the collaboration on a European level and would have to be implemented e.g. in the framework of a HorizonEurope project. Through the successful validation of our developed approach, it would be realistic to replace the current state-of-the- art method in forensic age estimation by a method that could be used with healthy subjects without the need for potentially harmful ionizing radiation.