Interpretable AI for Lung Cancer Screening
Published:Dec 31, 2025 00:23
•1 min read
•ArXiv
Analysis
This paper addresses the limitations of current lung cancer screening methods by proposing a novel approach to connect radiomic features with Lung-RADS semantics. The development of a radiological-biological dictionary is a significant step towards improving the interpretability of AI models in personalized medicine. The use of a semi-supervised learning framework and SHAP analysis further enhances the robustness and explainability of the proposed method. The high validation accuracy (0.79) suggests the potential of this approach to improve lung cancer detection and diagnosis.
Key Takeaways
- •Proposes a radiological-biological dictionary to link radiomic features with Lung-RADS semantics.
- •Employs a semi-supervised learning framework to improve generalizability.
- •Utilizes SHAP analysis to identify key radiomic features corresponding to Lung-RADS semantics.
- •Achieves a validation accuracy of 0.79, demonstrating the potential for improved lung cancer screening.
Reference
“The optimal pipeline (ANOVA feature selection with a support vector machine) achieved a mean validation accuracy of 0.79.”