Habitat Model Based on Lung CT for Predicting Brain Metastasis in Patients with Non-Small Cell Lung Cancer

Area of Science:

• Radiomics and Medical Imaging
• Oncology
• Computational Pathology

Background:

• Brain metastasis (BM) in lung cancer is linked to primary tumor heterogeneity.
• Predicting BM risk is crucial for non-small cell lung cancer (NSCLC) patient management.
• Enhanced CT imaging offers potential for non-invasive tumor characterization.

Purpose of the Study:

• To develop and validate a habitat-based radiomics model for predicting BM risk in NSCLC.
• To assess the model's performance against a whole-tumor radiomics approach.
• To investigate the role of subregional tumor heterogeneity in BM development.

Main Methods:

• Retrospective analysis of 195 NSCLC patients with enhanced CT lung imaging.
• Segmentation of tumors into subregions using k-means clustering (intensity, entropy).
• Radiomics feature extraction and selection; logistic regression models (whole-tumor vs. habitat-based).

Main Results:

• The habitat-based radiomics model achieved an AUC of 0.819, outperforming the whole-tumor model (AUC 0.728).
• The habitat-based model showed significantly superior predictive performance (p=0.022).
• Subregional heterogeneity analysis was key to the improved prediction accuracy.

Conclusions:

• Habitat-based radiomics models effectively predict BM risk in NSCLC.
• Analyzing subregional tumor heterogeneity is vital for improving predictive accuracy.
• This approach enhances risk stratification for NSCLC patients.