PaCL: Patient-aware contrastive learning through metadata refinement for generalized early disease diagnosis

Vandan Gorade, Sparsh Mittal, Rekha Singhal

Abstract: Highlights•Patient-aware contrastive learning using metadata such as age, gender, etc.•Use of an inter-class separability objective and an intra-class diversity objective.•Results on six real-world medical imaging tasks spanning ophthalmology, radiology, and dermatology images.•Extensive comparison is performed with the previous work•Results are presented under both supervised and semi-supervised setting