Radiomics-based machine learning to evaluate immunotherapy efficacy in non-small cell lung cancer patients with bone...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Assessing treatment response in non-small cell lung cancer (NSCLC) bone metastases, especially with immune checkpoint inhibitors (ICIs), is challenging.
• Current response criteria are not optimized for bone metastases, leading to inconsistent evaluations.

Purpose of the Study:

• To develop and validate a radiomics-based machine learning (ML) framework for non-invasively distinguishing immunotherapy response categories (progression, stable disease, partial response) in NSCLC patients with bone metastases.

Main Methods:

• Chest CT scans from 99 NSCLC patients before and during ICI therapy were analyzed.
• Radiomic features were extracted from segmented bone structures, and ML classifiers (random forest, XGBoost, SVM) were trained and optimized using feature selection pipelines.
• Model performance was evaluated using AUC, F1-score, accuracy, sensitivity, and specificity.

Main Results:

• Post-treatment radiomic features showed superior performance.
• The random forest model achieved an AUC of 0.94, F1-score of 0.79, accuracy of 0.79, sensitivity of 0.80, and specificity of 0.83.
• Clinical features did not significantly improve model performance, and models using overall response outperformed those based on the largest lesion.

Conclusions:

• Post-treatment CT radiomics effectively capture therapy-induced skeletal changes in NSCLC bone metastases.
• This radiomics framework enables differentiation of immunotherapy response categories non-invasively.
• The findings support radiomics as a valuable tool for response assessment and personalized treatment strategies in NSCLC patients with bone metastases.