Radiomics based Lung cancer staging

Radiomics based Lung cancer staging classifies lung cancer stage from CT-derived radiomics features to support TNM-guided staging and treatment stratification.

Key Features:

• Input data: Uses lung computed tomography (CT) image series as the primary input.
• Datasets: Utilizes open datasets including NSCLC-Radiomics and NSCLC-Radiogenomics for development and testing.
• Radiomics feature extraction: Calculates radiomics features from segmented tumor and lung volumes.
• Segmentation targets: Extracts features from both tumor-specific regions and lung lobe volumes.
• Feature harmonization: Combines tumor-specific and lung lobe radiomics features to enhance classification performance.
• Classification scheme: Implements a binary classifier with Class '0' representing overall stages I and II and Class '1' representing overall stages III and IV.
• Guidelines alignment: Staging is guided by the Tumour-Node-Metastasis (TNM) system and mapped according to NCCN-based rules.
• Modeling and validation: Applies various feature selection and classification methods validated through multiple cross-validation folds and external testing splits.
• Performance: Reported average precision of 77.5% and recall of 78.7% across validation experiments.

Scientific Applications:

• Stage stratification for treatment planning: Stratifies patients into lower (I–II) versus higher (III–IV) stage groups to inform treatment decisions.
• Case prioritization: Enables prioritization of lung cancer cases for clinical review based on predicted stage.
• Clinical research and federated AI: Supports AI development and validation within initiatives such as the INCISIVE federated image repository for cancer research.
• Objective radiomics-based staging: Provides reproducible radiomics measurements to assist comparative studies and external validation of staging models.

Methodology:

Data acquisition from NSCLC-Radiomics and NSCLC-Radiogenomics CT series; radiomics feature calculation from segmented tumor and lung volumes; feature harmonization by combining tumor-specific and lung lobe features; model training using various feature selection and classification methods with validation via multiple cross-validation folds and external testing splits; binary classification mapped to stages I–II versus III–IV per NCCN/TNM guidance.