Masked Attention: A Novel Approach to Enhance Interpretability of Vision Transformers in Computational Pathology

Masking background patches in the attention mechanism of Vision Transformers enhances model interpretability without compromising performance in computational pathology tasks.