AI-Powered Precision Diagnosis of Thyroid Anomalies in Ultrasound Scans

Abstract

The increasing global prevalence of thyroid cancer has created an urgent need for enhanced diagnostic precision in thyroid anomaly detection through ultrasound scans. Manual interpretation of thyroid nodules in ultrasound images often poses challenges for radiologists due to factors such as overlapping tissues, low image contrast, and the presence of small or indistinct nodules. In response to this diagnostic complexity, an AI-powered computer-aided diagnosis (CAD) system has been developed to automate and refine the identification of thyroid anomalies. The system is designed to significantly aid radiologists in making timely and accurate decisions by leveraging advanced deep learning methodologies.A publicly accessible thyroid ultrasound imaging dataset was employed for model training, validation, and evaluation. The dataset encompasses a diverse array of sonographic thyroid nodules, including benign and malignant samples. To effectively detect and classify these nodules, object detection algorithms were implemented, with particular focus on the YOLO (You Only Look Once) framework. The model variant YOLOv5x6 was chosen due to its capacity to perform high-speed and high-precision detection. Multiple optimization strategies were incorporated to improve performance: enhanced feature extraction for learning complex patterns, data augmentation for better generalization, and class imbalance rectification using oversampling and weighted loss functions. In addition, the model architecture was fine-tuned to boost sensitivity in identifying small and overlapping nodules.Among the evaluated models, YOLOv5x6 achieved superior performance with a precision of 0.561, recall of 0.835, and a mean average precision (mAP) of 0.650. These results highlight the robustness and reliability of the system in detecting thyroid nodules across varied sonographic conditions. The integration of artificial intelligence into medical imaging workflows demonstrates the potential to accelerate diagnostic processes, minimize oversight, and enhance overall clinical outcomes. This diagnostic tool offers a dependable solution that aligns with the growing demand for accurate, efficient, and automated thyroid anomaly detection in the healthcare domain.