Related Experiment Video
Updated: Sep 11, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.6K
Cost-Efficient Early Diagnostic Tool for Lung Cancer: Explainable AI in Clinical Systems.
Anu Maria Sebastian1,2, David Peter2, T P Rajagopal3
1Department of Computer Science and Engineering, Indian Institute of Information Technology, Kottayam, Kerala, India.
Technology in Cancer Research & Treatment
|August 14, 2025
Summary
This study introduces an explainable artificial intelligence (XAI) model using clinical data for safer, more affordable lung cancer diagnosis. The model shows strong performance and improves detection of rare cases, offering a promising pre-screening tool.
Area of Science:
- Oncology
- Medical Informatics
- Artificial Intelligence
Background:
- Lung cancer has the highest global mortality rate, often due to delayed diagnosis and treatment.
- Current diagnostic methods like radiomics carry health risks, and advanced alternatives are costly and inaccessible.
- Machine learning and artificial intelligence offer safer, more inclusive, and affordable diagnostic solutions, but lack clinical interpretability.
Purpose of the Study:
- To develop a safe, inclusive, and cost-effective lung cancer diagnostic method using explainable artificial intelligence (XAI) and routine clinical data.
- To enhance the diagnostic performance and interpretability of AI models for lung cancer detection.
- To address the challenge of diagnosing rare lung cancer cases and improve clinician adoption of AI tools.
Main Methods:
- Developed an XAI model using a stacking ensemble of Artificial Neural Network (ANN) and Deep Neural Network (DNN).
- Incorporated rare medical cases using Adaptive Synthetic Sampling (ADASYN) to improve detection of challenging diagnoses.
- Utilized SHapley additive exPlanations (SHAP) for model interpretability and counterfactual explanations to identify misdiagnosis factors.
Main Results:
- The XAI model achieved high performance: 0.8558 accuracy, 0.8600 AUC, 0.8092 precision, 0.9282 recall, and 0.8646 F1-score.
- Rare case detection improved by over 50%, demonstrating enhanced ability to identify challenging diagnoses.
- SHAP analysis identified key diagnostic features including Erythrocyte sedimentation rate (ESR), intoxication factors, hemoglobin, and neutrophil counts, revealing novel associations like tobacco use and ESR.
Conclusions:
- The developed XAI model demonstrates significant potential as a pre-screening tool for early lung cancer detection.
- The model's interpretability and improved rare case detection enhance its suitability for clinical support.
- Further development with larger, diverse datasets could lead to a robust and scalable AI-driven diagnostic solution for lung cancer.
