Counterfactual Analysis for Digital Histopathology Slides Using Human Interpretable Features

Hakim Benkirane, Maria Vakalopoulou, Stefan Michiels, Paul-Henry Cournède, William Lotter

Keywords: Histopathology, Counterfactual, Interpretability

Abstract: Recent advancements in deep learning techniques have greatly improved the precision and efficiency of computational pathology processes, facilitating diagnosis, outcome forecasting, and identification of genetic markers and disease progression. However, a significant challenge hindering the integration of these computational tools into clinical practice is the lack of interpretability of their results. In this paper, we propose a novel method for counterfactual analysis on histopathology slides to provide clear and understandable explanations based on human interpretable features for predictive tasks at the slide level. Our method addresses the challenge of generating interpretable explanations for high-dimensional tabular data in a computationally efficient manner, outperforming state-of-the-art methods by generating explanations approximately 10 times faster. This advancement holds promise for enhancing the adoption and effectiveness of deep learning models in clinical settings, ultimately improving patient care and outcomes.

Submission Number: 49