Thyroid nodules, characterized by abnormal growths within the thyroid gland, are frequently identified through ultrasound imaging and necessitate precise risk stratification to guide clinical management. The Thyroid Imaging Reporting and Data System (TI-RADS) provides a standardized framework for categorizing these nodules based on ultrasound features such as echogenicity, margins, shape, and calcification, thus assisting in determining the necessity for biopsy or further diagnostic procedures. However, the inherent limitations of manual interpretation within the TI-RADS system highlight the need for enhanced diagnostic methodologies. This paper introduces an advanced, AI-driven system that leverages state-of-the-art deep learning models-specifically the Segment Anything Model (SAM) and You Only Look Once (YOLO V8)-to automate and refine the classification of thyroid nodules. By seamlessly integrating TI-RADS criteria with machine learning capabilities, this system enables precise detection of malignancy-associated features, surpassing traditional approaches in sensitivity and specificity. Iterative model training and multi-modal feature extraction allow for a comprehensive risk assessment, addressing the ambiguities of human interpretation and limited datasets. Our results demonstrate significant improvements in diagnostic accuracy, offering a transformative approach to thyroid nodule evaluation and paving the way for more effective and scalable clinical solutions in thyroid cancer screening.