peritumoral regions

peritumoral regions extracts and evaluates computed tomography (CT) radiomic features from peritumoral regions surrounding lung cancer lesions to identify stable and reproducible metrics for prognostic modeling and clinical evaluation.

Key Features:

• Radiomic Feature Extraction: Systematically extracts 264 radiomic features from peritumoral zones at incremental distances of 3 to 12 mm outside tumor segmentations, including statistical, histogram, and texture-based metrics, using Image Biomarker Standardisation Initiative algorithms.
• Stability Assessment: Defines stability as feature consistency across different segmentations and evaluates it using the "moist run" dataset.
• Reproducibility Evaluation: Assesses feature consistency across image acquisitions using the Reference Image Database to Evaluate Therapy Response test-retest dataset.
• Concordance Correlation Coefficient (CCC): Quantitatively measures both stability and reproducibility using the Concordance Correlation Coefficient (CCC).
• Clinical Utility Testing: Applies stable and reproducible features to three previously published lung cancer datasets with overall survival as the endpoint to evaluate predictive utility.
• Feature Exclusion: Excludes unstable feature families such as Laws and wavelet texture features from certain peritumoral distances to refine feature selection.

Scientific Applications:

• Feature Selection Pipeline: Identifies a subset of stable and reproducible CT radiomic features to support feature selection and reduce overfitting in radiomic analyses.
• Model Reproducibility: Prioritizes features that demonstrate stability and reproducibility to enable generation of repeatable radiomic models.
• Clinical Decision Support: Supports integration of reliable peritumoral features into clinical decision-making workflows for patient stratification and treatment planning informed by overall survival models.

Methodology:

Extraction of 264 radiomic features at 3–12 mm peritumoral distances using Image Biomarker Standardisation Initiative algorithms; stability assessed across segmentations with the "moist run" dataset; reproducibility assessed across acquisitions using the Reference Image Database to Evaluate Therapy Response test-retest dataset; quantitative evaluation via Concordance Correlation Coefficient (CCC); exclusion of unstable Laws and wavelet texture features from specified peritumoral distances; application of selected features to three published lung cancer datasets with overall survival endpoints.