Effect of machine learning methods on predicting NSCLC overall survival time based on Radiomics analysis

Area of Science:

• Radiomics and Medical Imaging
• Machine Learning in Oncology
• Computational Pathology

Background:

• Non-small cell lung cancer (NSCLC) survival prediction remains a challenge.
• Radiomics analysis extracts quantitative features from medical images.
• Machine learning offers potential for improved prognostic models.

Purpose of the Study:

• To evaluate machine learning methods for predicting overall survival (OS) in NSCLC.
• To assess the efficacy of radiomics features in survival prediction.
• To identify optimal feature selection and machine learning models for NSCLC OS prediction.

Main Methods:

• Extracted 339 radiomic features from pretreatment CT scans of NSCLC patients.
• Quantified tumor characteristics including shape, size, intensity statistics, and textures.
• Investigated 5 feature selection and 8 machine learning methods, evaluating performance with concordance index.

Main Results:

• Gradient boosting linear models with concordance index feature selection achieved the best performance.
• Achieved a Concordance Index of 0.68 (95% CI: 0.62–0.74).
• Kaplan-Meier and log-rank tests were used for survival curve evaluation.

Conclusions:

• Machine learning and radiomics analysis show promise for accurate NSCLC OS prediction.
• Specific models demonstrate potential for clinical application in NSCLC prognosis.
• Further validation is warranted to confirm the predictive accuracy.