A novel case-based reasoning system for explainable lung cancer diagnosis

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Oncology

Background:

• Lung cancer is a major global health concern with higher survival rates upon early detection.
• Artificial intelligence (AI) decision support systems aid diagnosis but often lack transparency, hindering physician trust.
• Explainability is crucial for AI adoption in critical medical applications like cancer diagnosis.

Purpose of the Study:

• To develop an explainable case-based reasoning (XCBR) approach for lung cancer diagnosis.
• To enhance AI-driven medical decision support systems with transparent and trustworthy predictions.
• To improve early lung cancer detection through explainable AI.

Main Methods:

• Proposed an explainable case-based reasoning (XCBR) framework incorporating case complexity.
• Employed a hierarchical classification system with Naïve Bayes (NB) and Multilayer Perceptron (MLP).
• Integrated Shapley additive explanations (SHAP) for MLP transparency and Harris Hawks Optimization for feature selection.

Main Results:

• Achieved high diagnostic accuracies of 94.47% and 100% on two distinct datasets.
• Demonstrated comparable classification accuracy to state-of-the-art methods via Wilcoxon signed-rank test.
• Showcased superior explainability by prioritizing case complexity in predictions.

Conclusions:

• The proposed XCBR approach offers a trustworthy and explainable AI solution for lung cancer diagnosis.
• Enhanced explainability can improve physician adoption and confidence in AI diagnostic tools.
• This method is particularly suitable for complex and serious diseases requiring transparent decision-making.