An Explainable Computer Vision in Histopathology: Techniques for Interpreting Black Box Model
Subrata Bhattacharjee, Yeong-Byn Hwang, Kobiljon Ikromjanov, Rashadul Islam Sumon, Hee‐Cheol Kim, Heung‐Kook Choi · 2022
Computer vision is a field of artificial intelligence (AI) that is being used increasingly in histopathology to identify pathologies in slide images with a high degree of accuracy. In this paper, we focus on the different interpreting techniques of explainable computer vision (XCV). Analysis of histopathology images is a challenging task, and specialized knowledge is mandatory to make AI decisions. To carry out this analysis, a deep learning model has been used to classify and differentiate the scoring (i.e., benign and malignant) of Prostate cancer (PCa). However, the AI models are complex and opaque, and it is important to understand model decision-making. Therefore, to address this problem, we present three techniques for accountability and transparency of the model, namely Activation Layer Visualization (ALV), Local Interpretable Model-Agnostic Explanation (LIME), SHapley Additive exPlanations (SHAP), and Gradient-weighted Class Activation Mapping (Grad-CAM). XCV is AI in which the results of the black-box model can be understood by humans. The robustness of our model has been confirmed by using an external test dataset including 100 histopathology images. The model performance has been evaluated using the receiver operating characteristic (ROC) curve.