Adaptive knowledge distillation for invasive ductal carcinoma grading using histopathological images

Invasive Ductal Carcinoma Grading (IDC) through histopathological image analysis is a critical task, requiring high accuracy and operational efficiency. Knowledge distillation (KD), an advanced technique in which a smaller model learns from a larger model, represents a promising solution for balancing performance and computational efficiency. In this paper, a novel KD-based technique is developed to efficiently classify IDC grading histopathology images. Unlike traditional distillation, our method proposed a KD algorithm consisting of teacher response supervision as well as intermediate representations supervision, along with an adaptive weighting mechanism. To our knowledge, this study represents the first application of KD for IDC grading. The findings demonstrate that our proposed KD significantly improves model efficiency while maintaining high classification accuracy.