RareSim

RareSim is a research project focused on improving the diagnosis and management of rare diseases with the help of advanced AI technologies. In RareSim, we propose gathering data from our clinical practice partners, the Children’s Hospital Zurich (KISPI) and the University Hospital Zurich (USZ), to develop novel approaches - particularly similarity functions - for automatically determining the similarity between any two patients. These functions are subsequently used to identify patients harboring rare diseases. Computationally, this corresponds to a classification problem, in which a computer assigns an ORPHA code to patients based on their similarity to other patients with known rare disease, i.e., known ORPHA codes. The first goal of the next phase of the project is to develop a series of similarity functions and benchmark them on real-world data against measures from the literature.

A key feature of RareSim is its human-in-the-loop approach, which means it actively involves clinicians in the AI learning process. We aim to incorporate input and feedback from clinicians to improve the computation of the similarity functions. To do so, our second goal this year is to develop a patient data exploration interface to provide clinicians with a comparative view of two patient cases, while also allowing us to gather additional relevant information input from the clinicians. Subsequently, this would help us improve our similarity computation and make better predictions.