Improving thyroid cancer diagnosis through stacked explainable machine learning integrating ultrasound and Raman cytological features

Indeterminate thyroid cytology remains a diagnostic challenge. We investigated whether ultrasound-derived clinical features and Raman spectra acquired from fine-needle aspiration samples could be integrated within an explainable multimodal machine learning framework. The analytical cohort comprised 203 patients from a single center who underwent thyroid surgery, including 64 follicular adenomas (FA) and 139 malignant lesions according to postoperative histology. The malignant class comprised 117 papillary thyroid carcinomas (PTC), 21 follicular-variant papillary thyroid carcinomas (FV-PTC), and one mixed FA+PTC lesion. The fixed primary architecture combined a clinical Logistic Regression classifier and a raw-Raman Extra Trees classifier through a Logistic Regression meta-learner. Internal validation used stratified fivefold cross-validation repeated 20 times, with all data-dependent preprocessing, stacking, and operating-threshold selection restricted to the corresponding training data. The stacked model achieved a mean repeated out-of-fold AUC of 0.856 (95% case-level bootstrap CI 0.796–0.909; resampling SD 0.011). In an exploratory held-out analysis of 93 TIR3A/TIR3B cases, the corresponding AUC was 0.859 (95% CI 0.783–0.922; resampling SD 0.019). SHAP analyses identified model-dependent clinical and Raman associations that were stable at the clinical-feature and Raman-region levels across resampled outer folds. The multimodal model showed promising internal discrimination in a single-center, surgery-selected cohort. Prospective multicenter validation of a locked model and a prespecified clinical decision strategy is required before assessing clinical utility or any effect on surgical management.