Human-Centered Explainable AI Framework and Its Application on Detection of Invasive Ductal Carcinoma from Breast Histopathology Images

Abstract

In healthcare, AI adoption requires high predictive accuracy, combined with explainability, factoring in the human element to build trust among clinicians. The authors introduce a novel Human-Centered Explainable AI (HCXAI) paradigm for the analysis of breast histopathology images.Convolutional Neural Network (CNN) models have been developed as a result of the Camelyon-16 challenge in 2016 to aid radiologists and other physicians for diagnosing breast cancer. Ductal Carcinoma in Situ (DCIS), which is the medical term for early-stage breast cancer, has received particular attention. Numerous big businesses have invested in research on this topic, including Google Deepmind, which in 2020 built a breast cancer diagnostic model that has been shown to be more accurate than radiologists themselves. In this work, popular classifiers such as CNN, Support Vector Machine (SVM), Random Forest were used in order to correctly make prediction and classify the Invasive Ductal Carcinoma. Explainable Artificial Intelligence (XAI) was extensively used in this work to further extract interpretability of images and analyze different prediction models and their performances in terms of different parameters.