Attention-Based Deep Multiple Instance Learning With Adaptive Instance Sampling

One challenge of training deep neural networks with giga-pixel whole-slide images (WSIs) in computational pathology is the lack of annotation at the pixel level or smaller image regions. Manual annotation of image regions by an expert is costly and time-consuming. Multiple instance learning (MIL) and its attention-based versions are typical weakly supervised learning methods dealing with slide labels instead of image regions and reducing the cost of annotation. However, training a deep neural network with thousands of image regions per slide is computationally expensive and needs a lot of time for convergence. This paper proposes a fast adaptive attention-based deep MIL approach. It adaptively sub-selects image regions that are highly predictive of outcome and ignores image regions with little or no information. We empirically show that our proposed approach outperforms the random sampling while it is faster than the standard attention-based MIL method.