Aims: Current TI-RADS is constrained by subjective, categorical assessments such as "irregular margins" and "taller-than-wide", leading to inter-observer variability and lack of continuous stratification. We aim to translate these into objective, quantitative morphological signatures using radiomics. By mapping margins to sphericity and shape to elongation, we sought to improve diagnostic accuracy and safely reduce the volume of unnecessary fine-needle aspirations (FNA). Methods: In the retrospective open benchmark ThyroidXL dataset of 3,354 nodules (n=877, 26% malignant), we compared two risk stratification models. 1.Baseline model: logistic regression utilising standard TI-RADS scores combined with the largest nodule dimension. TI-RADS was retrained to address the issue of multiple sub-centimeter cancer nodules in this dataset which resulting in 40% sensitivity for native TI-RADS. 2.Experimental morphological model: a non-linear Support Vector Machine (SVM) trained exclusively on a 2-feature signature - sphericity and elongation. Results: The morphological model significantly outperformed the baseline (AUC 0.97 versus 0.91). To assess real-world clinical utility, models were evaluated at a standardised 90% sensitivity cutoff. The standard TI-RADS based approach required 1.6 FNAs to identify a single true positive malignancy. In contrast, the quantitative morphological model required only 1.2 FNAs per true positive, demonstrating a marked reduction in unnecessary invasive investigations for benign disease. Conclusion: Replacing subjective binary features with computer-aided, continuous morphological quantification enhances thyroid nodule risk assessment. This interpretable 2-feature model demonstrates superior performance over optimized TI-RADS, accurately detecting even challenging sub-centimeter malignancies whilst reducing the clinical burden and patient morbidity associated with diagnostic FNAs.