Development of tirads scoring system based on artificial intelligence for differential diagnosis of thyroid nodules

Patients with thyroid conditions, initially evaluated by a clinician such as an internal medicine specialist, are subsequently referred to radiology departments for detailed examination. This process accounts for approximately 10% of the ultrasonography workload; it corresponds to 20% of the interventional radiology workload. Thyroid nodules are a widespread health issue affecting a large portion of the population, making accurate and timely diagnosis crucial. However, the subjectivity and human errors inherent in traditional diagnostic methods can negatively impact diagnostic accuracy. This thesis aims to automatically evaluate thyroid nodules from ultrasound images using deep learning techniques. A new dataset specific to thyroid nodules has been developed, and artificial intelligence models were created using this dataset, enabling the detection of nodule boundaries and TIRADS features. As a result, this study demonstrates the feasibility of AI-supported systems in thyroid cancer screening and provides a tool that can enhance clinical decision-making processes.