Explainable AI for Breast Cancer Detection: A LIME-Driven Approach

Artificial Intelligence is transforming the healthcare industry due to the increasing accessibility of organized and unorganized information and the rapid development of analytical techniques. As artificial intelligence becomes more significant in healthcare, concerns are arising regarding the lack of transparency, explainability, and the possibility of bias in model predictions. The goal of this paper is to utilize interpretable machine learning to provide a better understanding of breast cancer using the Local Interpretable Model-agnostic Method (LIME). This study uses Local Interpretable Model-Agnostic Explanations to explain how the machine-learning model accurately classifies breast cancer cases as either Benign or Malignant. It provides a LIME plot for Benign cases, highlighting the significant role of "Bare Nuclei" where lower values strongly suggest Benign predictions. Other features like Normal Nucleoli, Marginal adhesion, single adhesion, Mitoses, and uniformity of cell size also contribute to the Benign class prediction when they fall below a particular threshold. The study also presents a LIME plot for Malignant cases, emphasizing the importance of "Bare Nuclei" and Clump thickness, where higher values indicate a higher likelihood of Malignant predictions. Other features like Normal Nucleoli, Marginal adhesion, Bland chromatin, uniformity of cell size, and Mitoses also contribute to Malignant predictions when their values exceed specific thresholds. The feature "Concave points_worst" influences the model’s benign predictions when it is below 0.07, while the "texture" feature affects predictions when it exceeds 29.41. Furthermore, the study provides further explanations for Malignant predictions based on "Concave points_worst" and "Texture." These findings provide valuable insights into the decision-making process of the model, making it more interpretable and useful for breast cancer diagnosis.