B-Cos Aligned Transformers Learn Human-Interpretable Features
This work addresses the problem of interpretability in AI models for medical experts in computational pathology, offering a novel method that enhances both transparency and performance.
The paper tackles the lack of interpretability in Vision Transformers for computational pathology by introducing B-cos Vision Transformers (BvT) and B-cos Swin Transformers (Bwin), which replace linear transformations with B-cos transforms to promote weight-input alignment. In a blinded study, medical experts ranked BvTs above ViTs for capturing biomedically relevant structures, and Bwin improved the F1-score by up to 4.7% on two public datasets compared to Swin Transformers.
Vision Transformers (ViTs) and Swin Transformers (Swin) are currently state-of-the-art in computational pathology. However, domain experts are still reluctant to use these models due to their lack of interpretability. This is not surprising, as critical decisions need to be transparent and understandable. The most common approach to understanding transformers is to visualize their attention. However, attention maps of ViTs are often fragmented, leading to unsatisfactory explanations. Here, we introduce a novel architecture called the B-cos Vision Transformer (BvT) that is designed to be more interpretable. It replaces all linear transformations with the B-cos transform to promote weight-input alignment. In a blinded study, medical experts clearly ranked BvTs above ViTs, suggesting that our network is better at capturing biomedically relevant structures. This is also true for the B-cos Swin Transformer (Bwin). Compared to the Swin Transformer, it even improves the F1-score by up to 4.7% on two public datasets.