Predictive Modelling Using Thyroid Cartilage Segmentation and Radiomic Features: A Feasibility Study

Laryngeal cancer, one of the top three head and neck cancers, requires timely diagnosis and staging for effective management and improved patient outcomes. Thyroid cartilage penetration indicates advanced cancer and is crucial for treatment planning. However, identifying cartilage abnormalities on CT images is challenging due to age-related changes, and use of machine learning (ML) models has been proposed as a possible way forward. In this feasibility study, we manually segmented thyroid cartilage in 39 CT images from the HaN-Seg dataset using 3D Slicer. Radiomic features were extracted with Slicer Radiomics, and statistical and ML analyses were conducted using Jamovi and MATLAB. Manual segmentation of thyroid cartilage was successful, yielding 107 radiomic features. Significant gender and age-related differences were identified. ML models classified gender with 100% accuracy and age group with 85.71% accuracy. Regression models showed improved accuracy with transformed variables. Radiomic analysis of thyroid cartilage is promising for classifying age-related change. Subsequent studies on this could aid in laryngeal cancer staging, distinguishing between normal and tumour-infiltrated cartilage.