XtraTracks 2024 - GestaltMatcher
Next-generation phenotyping for rare genetic disorders – 22. Mai 2024
Kurzbeschreibung
Worldwide, rare genetic disorders affect more than 6.2% of the population. The long diagnostic process is often called the ‘diagnostic odyssey.’ With the recent advances in computer vision, many next-generation phenotyping (NGP) approaches such as DeepGestalt have shown a strong ability to differentiate rare disorders and are widely used by clinicians in clinics. However, the current NGP approaches for rare disorders still have limitations on two aspects: current approaches do not support ultra-rare and novel disorders and no publicly available dataset. Therefore, we proposed GestaltMatcher and GestaltMatcher Database (GMDB) to tackle the current difficulties.
We first developed GestaltMatcher as an extension to DeepGestalt to support ultra-rare and novel disorders. GestaltMatcher first encoded the frontal image into a 320-dimensional Facial Phenotype Descriptor (FDP). We further formed a Clinical Face Phenotype Space by the FDPs and quantified the facial syndromic similarities among the patients by calculating the cosine distance between two FDPs in the space. This approach can support ultra-rare disorders and novel diseases and analyze the patients’ similarities to explore the novel gene-phenotype relationship.
To address the problem of lacking a public medical image dataset, we proposed GMDB, an open-access medical image database. GMDB hosts the medical images curated from the publication and the consented patients, providing a valuable resource for deep learning purposes and serving as reference material for clinician-scientists to easily access and analyze medical images.
In the end, we envision that GestaltMatcher and GMDB can be integrated into a diagnostic platform and further connected with patient match platforms such as MatchMaker Exchange to enable global collaboration and further improve the diagnosis of rare genetic disorders.
Vortragende
Jean Tori Pantel (RWTH Aachen) und Dr. Tzung-Chien Hsieh (UK Bonn)
Vortragstermin
Datum: 22. Mai 2024
Beginn: 17:00 Uhr
Online: Veranstaltungsraum