National Library of Medicine 5R01LM013766-04

Multi-modal unsupervised embeddings to advance machine learning in healthcare

This award funds the development of unsupervised machine learning methods to create multi-modal embeddings from diverse biomedical data to enhance healthcare predictive models, including federated learning across institutions to reduce biases and improve scalability.

Description

Develop novel methods based on unsupervised machine learning to derive low-dimensional vector-based representations of medical concepts and patient clinical histories from large-scale, multi-modal, and domain-free biomedical datasets to improve scalability, generalizability, and effectiveness of machine learning models in healthcare. The project includes creating multi-modal embeddings using heterogeneous EHRs, linked biobanks, and electrocardiogram waveform data from five hospitals within the Mount Sinai Health System in New York, NY, and publicly available medical knowledge, developing a scalable framework to compute these embeddings, and a federated learning system to share and combine embeddings across institutions. The embeddings will be applied to EHR-based disease phenotyping, onset prediction, and subtyping, with extendable approaches to other data types such as clinical images.

View original record