A robust and lightweight deep attention multiple instance learning algorithm for predicting genetic alterations

Bangwei Guo, Xingyu Li, Miaomiao Yang, Hong Zhang, Xu Steven Xu

Abstract: Highlights•The lightweight AMIML model allows small-scale self-attention and quicker turnaround compared to standard self-attention models.•AMIML shows excellent robustness and outperforms other existing deep learning and machine learning models.•AMIML improves the prediction of a wide range of genes over the current literature reports.•The interpretability of AMIML workflow can help pathologists to zero in on important ROIs