Measuring and Predicting Where and When Pathologists Focus their Visual Attention while Grading Whole Slide Images of Cancer

TL;DR

This study develops a transformer-based model to predict the dynamic visual attention patterns of pathologists viewing prostate cancer slides, aiming to enhance pathology training and decision support systems.

Contribution

We introduce a novel two-stage transformer model that predicts both static attention heatmaps and dynamic scanpaths during whole slide image analysis.

Findings

Our model outperforms chance and baseline models in predicting attention scanpaths.

The approach effectively captures the spatio-temporal attention behavior of pathologists.

Tools from this work could improve pathology training by modeling expert attention patterns.

Abstract

The ability to predict the attention of expert pathologists could lead to decision support systems for better pathology training. We developed methods to predict the spatio-temporal (where and when) movements of pathologists' attention as they grade whole slide images (WSIs) of prostate cancer. We characterize a pathologist's attention trajectory by their x, y, and m (magnification) movements of a viewport as they navigate WSIs using a digital microscope. This information was obtained from 43 pathologists across 123 WSIs, and we consider the task of predicting the pathologist attention scanpaths constructed from the viewport centers. We introduce a fixation extraction algorithm that simplifies an attention trajectory by extracting fixations in the pathologist's viewing while preserving semantic information, and we use these pre-processed data to train and test a two-stage model to predict the dynamic (scanpath) allocation of attention during WSI reading via intermediate attention heatmap prediction. In the first stage, a transformer-based sub-network predicts the attention heatmaps (static attention) across different magnifications. In the second stage, we predict the attention scanpath by sequentially modeling the next fixation points in an autoregressive manner using a transformer-based approach, starting at the WSI center and leveraging multi-magnification feature representations from the first stage. Experimental results show that our scanpath prediction model outperforms chance and baseline models. Tools developed from this model could assist pathology trainees in learning to allocate their attention during WSI reading like an expert.