Robustness of Radiomics in Pre-Surgical Computer Tomography of Non-Small-Cell Lung Cancer

Area of Science:

• Medical Imaging
• Radiology
• Oncology

Background:

• Radiomic features are increasingly utilized in computed tomography (CT) for non-small cell lung cancer (NSCLC) analysis.
• The robustness of these radiomic features against variations in image segmentation remains a critical, undemonstrated aspect.
• Understanding segmentation variability is essential for reliable radiomic feature application in clinical practice.

Purpose of the Study:

• To evaluate the agreement and robustness of radiomic features across different segmentation methods in NSCLC CT.
• To assess the sensitivity of radiomic features to variations in segmentation parameters.
• To identify reproducible radiomic features for accurate NSCLC characterization.

Main Methods:

• Retrospective analysis of CT scans from 48 NSCLC patients.
• Manual delineation of three-dimensional regions of interest (3D-ROIs) by two expert radiologists.
• Intra-class correlation coefficient (ICC) used to assess feature agreement; sensitivity analysis performed on parameters like bin-width and pixel-distance.

Main Results:

• Shape features exhibited excellent agreement (ICC > 0.9) and minimal sensitivity to segmentation parameters.
• First-order and second-order features showed variable agreement, with a subset demonstrating excellent reproducibility and low parameter sensitivity.
• Reproducibility was also assessed concerning interpolator and isotropic resolution.

Conclusions:

• A select group of radiomic features demonstrate high reproducibility in NSCLC CT imaging.
• Shape features are particularly robust against segmentation variability.
• These findings support the potential for using a limited set of reliable radiomic features in clinical NSCLC assessment.