Improved Multisource Image-Based Diagnostic for Thyroid Cancer Detection: ANTHEM National Complementary Plan Research Project

Thyroid nodule evaluation relies heavily on ultrasound imaging, yet it suffers from significant inter-operator variability. To address this, we present a preliminary validation of the Synergy-Net platform, an AI-driven Computer-Aided Diagnosis (CAD) system designed to standardize acquisition and improve diagnostic accuracy. The system integrates a U-Net architecture for anatomical segmentation and a ResNet-50 classifier for lesion characterization within a Human-in-the-Loop (HITL) workflow. The study enrolled 110 patients (71 benign, 39 malignant) undergoing surgery. Performance was evaluated against histopathological ground truth. The system achieved an Accuracy of 90.35% (95% CI: 88.2–92.5%), Sensitivity of 90.64% (95% CI: 87.9–93.4%), and an AUC of 0.90. Furthermore, the framework introduces a multimodal approach, performing late fusion of imaging features with genomic profiles (TruSight One panel). While current results validate the 2D diagnostic pipeline, the discussion outlines the transition to the ANTHEM framework, incorporating future 3D volumetric analysis and digital pathology integration. These findings suggest that AI-assisted standardization can significantly enhance diagnostic precision, though multi-center validation remains necessary.