Validation report 004: Breast Cancer Detector Model
Published in Explainable Machine Learning 2023/2024 course, 2024
This study conducts a Red Teaming analysis on the Breast Cancer Detector Model, a Convolutional Neural Network designed for predicting breast cancer from tissue scans. Using eXplainable Artificial Intelligence (XAI) techniques, we assess the model’s reliability, investigating influences from unintended artifacts and evaluating its generalization with out-of-distribution samples. Our aim is to uncover vulnerabilities and enhance the model’s robustness in clinical applications.
Performed analysis proved strength of the model. Various vulnerabilities we were able to find turned out to be very intricate, unlike to be revealed in day to day usage. We assess the uncertainty while classifying the sample as infected tissue, to be promising field to improve.
We can conclude that phenomena discovered during LIME and SHAP analysis are deeply connected to this problem. Also, we advise care in ensuring the quality of input data, as vulnerabilities involving data augmentation were exposed.
Link to original datasets: Breast Histopathology Images, BreCaHAD.
Recommended citation: Mikolaj Drzewiecki, Monika Michaluk. (2024). "Red Teaming analysis of the Breast Cancer Detector Model." Github: ModelOriented/CVE-AI.
Download Paper