Prof. Dr. Lars Linsen, Institut für Informatik

The multiscale imaging strategy central to this initiative imposes novel data analysis challenges. The high complexity of the acquired data results from their nature of being volumetric, time-varying, large, multiscale, and forming cohorts). Meeting these challenges requires basic research in the fields of image analysis, machine learning, and visualization. Machine learning will be used to uncover inherent relationships between patterns at multiple scales. An interactive visual approach supports the user-centric analysis of detected features.

The deliverable of this project will be generally applicable, effective, and efficient methods supporting the overall goal of multiscale data analysis.

Cross-border migration movements of people with undocumented dates of birth have led to a need for forensic age assessments when the age of these people is of legal significance. In Germany and numerous other countries the age limit of 18 has the greatest practical relevance. Only the assessment of the ossification of the medial clavicular epiphysis currently allows proof beyond reasonable doubt of the completion of the 18th year of life. Thin-slice CT is currently considered the method of choice for imaging the medial clavicular epiphysis. The main objective of the project is to considerably improve the validity and reliability of age assessments based on a CT scan of the clavicles. This includes as a basis the establishment of a first reference population optimised in terms of sample size and age distribution, and free of individuals with pathologies or medications that may affect bone maturation. Three independent experts will determine stages of clavicular ossification years according to established scales to maximize validity. The use of machine learning techniques will objectify the assessment of the ossification status of the medial clavicular epiphysis; it will also allow the search for new features applicable at older ages, not readily available for the human eye. Ratings based on machine learning methodology (both using established scales and new predictors) will be transformed into an open demonstrator for a clinical decision support system which might later on help medicolegal specialists outside of the few existing expert centers worldwide to perform legal age assessment with an objectively derivable diagnostic accuracy.

Clinical practice often focuses on the investigation of one single disease, while the health status of a human is much more complex and may depend on many factors. Recently, cohort studies have been introduced to investigate, in longitudinal studies, the health status of an entire population (the cohort) by capturing health record data, whole-body medical imaging data, personal data including socio-economical circumstances, and even genetic sequencing data. Given this large amount of heterogeneous data, there is a lack of proper tools for its multi-variate analysis. In this project, we propose novel interactive visual analysis methods for testing hypotheses, supporting the generation of new hypotheses, and investigating changes over time. The goal is to allow for the detection of risk or biomarkers and even genetic associations in a multi-variate setting.In the second funding period, the research conducted in the first funding period shall be enhanced in various aspects. We will put a particular focus on the time aspect in multi-dimensional heterogeneous data from longitudinal studies, the analysis of influencing factors, analyzing multi-dimensional heterogeneous data with missing entries, and analyzing sparse high-dimensional data from genome-wide association studies.Moreover, we would like to validate the effectiveness of the proposed analysis methods by performing comparative visual analyses of the multi-dimensional heterogeneous data from different cohort studies.