ALFREDO: Active Learning with FeatuRe disEntangelement and DOmain adaptation for medical image classification

Dwarikanath Mahapatra, Ruwan B. Tennakoon, Yasmeen George, Sudipta Roy, Behzad Bozorgtabar, Zongyuan Ge, Mauricio Reyes

Abstract: Highlights•Proposed ALFREDO (Active Learning with FeatuRe disEntangelement and DOmain adaptation).•Our feature disentanglement extracts domain and task-specific features for classification.•We use novel metrics to quantify sample informativeness across domains with domain shift.•Evaluated our method on 3 public medical imaging datasets, benchmarking against other methods.