Artificial Intelligence-Based Diagnostic Assistance System for Thyroid Nodules in Ultrasound Imaging

Abstract

Thyroid cancer is one of the most common malignant endocrine diseases, and its early diagnosis remains a major challenge in medical imaging. Although thyroid ultrasound is the reference examination due to its non-invasive and accessible nature, its interpretation is affected by low image contrast, speckle noise, nodule variability, and operator dependency. To address these limitations, this work proposes an automatic thyroid nodule analysis system based on Artificial Intelligence (AI) and integrated into a Computer-Aided Diagnosis (CAD) framework. The proposed approach combines nodule segmentation and benign/malignant classification from ultrasound images. Experiments were conducted on public thyroid ultrasound datasets using preprocessing, data augmentation, and optimization techniques. The obtained results achieved a Dice Score of 85.44% and an IoU of 75.45% for segmentation, while classification reached an accuracy of 97.03% and an AUC of 99.05%. These results demonstrate the effectiveness of the proposed system and its potential to support clinical decision-making in thyroid cancer diagnosis.