Related Experiment Video
Updated: Jun 12, 2026
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Robust Semi-Supervised CT Radiomics for Lung Cancer Prognosis: Cost-Effective Learning with Limited Labels and SHAP
IEEE Transactions on Bio-Medical Engineering
|June 10, 2026
Summary
Semi-supervised learning (SSL) significantly improves lung cancer survival prediction from CT scans compared to supervised learning (SL), even with limited labeled data. This AI approach enhances accuracy and generalizability for clinical readiness.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
- Radiomics and Computational Pathology
Background:
- Computed tomography (CT) is crucial for lung cancer (LCa) management and AI-driven prognosis.
- Supervised learning (SL) models for LCa prognosis require extensive labeled data, which is often scarce in real-world scenarios.
Purpose of the Study:
- To develop and evaluate a semi-supervised learning (SSL) framework for improving lung cancer survival prediction using CT imaging.
- To assess the performance of SSL compared to SL under varying data availability conditions.
- To enhance the interpretability and clinical readiness of AI models for LCa prognosis.
Main Methods:
- Analyzed CT scans from 977 patients across 12 datasets, extracting 1,218 radiomics features.
- Implemented a semi-supervised learning (SSL) framework with pseudo-labeling using both labeled and unlabeled cases.
- Benchmarked 27 classifiers and assessed model sensitivity across scenarios with varying labeled and unlabeled data, using SHapley Additive exPlanations (SHAP) for interpretability.
Main Results:
- SSL outperformed SL in all metrics, improving overall survival prediction accuracy by up to 17%.
- The top SSL model achieved 0.90±0.01 accuracy in cross-validation and 0.88±0.01 externally, demonstrating robust performance with as little as 10% labeled data.
- SHAP analysis confirmed enhanced feature discriminability with SSL and provided insights into model decision-making.
Conclusions:
- An interpretable and cost-effective SSL framework is proposed for CT-based LCa survival prediction.
- The SSL framework enhances performance, generalizability, and clinical readiness through SHAP explainability and efficient unlabeled data utilization.