Semi-supervised medical image classification via distance correlation minimization and graph attention regularization

Abel Díaz Berenguer, Maryna Kvasnytsia, Matías Nicolás Bossa, Tanmoy Mukherjee, Nikos Deligiannis, Hichem Sahli

Abstract: Highlights•Novel semi-supervised learning (SSL) method for medical image classification.•Investigate distance correlation minimization to harness unlabeled data in SSL.•Graph-attention regularization to model affinities within the unlabeled data.•Experiments on four data sets comprising 2D and 3D images from different modalities.•Our SSL method excels in realistic single- and multi-label classification scenarios.