Automatic Discovery of Disease Subgroups by Contrasting with Healthy Controls

Researchers proposed Deep UCSL, a contrastive subgroup discovery method for biomedical data.

It uses healthy controls, a deep feature extractor, and EM-based optimization to learn disease-focused patient clusters.

The approach aims to filter out non-pathology variation and produce more interpretable subtypes.

Experiments on synthetic data and medical imaging benchmarks showed stronger subgroup quality than prior methods.