A study on ultrasound imaging for thyroid detection and classification using machine learning and deep learning techniques.

Thyroid diseases are increasingly prevalent and have a major influence on endocrine and metabolic functions. Ultrasound imaging is popular for early diagnosis and categorizing thyroid abnormalities because it is non-invasive, inexpensive, and real-time. From this paper, we planned to study the recent developments in computer-aided diagnosis (CAD) systems in detecting and classifying thyroid diseases based on ultrasound images. Researchers are focused on traditional image processing algorithms, machine learning techniques, and deep learning techniques to improve accuracy. Modern classifiers like convolutional neural networks (CNNs) have garnered much attention in nodule segmentation, benign/malignant classification, and feature extraction. Our study encompasses major methodologies, datasets, performance measures, and challenges. Our investigation seeks to integrate a comprehensive view of existing studies in this area, point out existing gaps, and make recommendations on future research in thyroid ultrasound image analysis. From our survey, we concluded that the Deep learning models have shown superior performance relative to traditional techniques in several benchmark studies with very good accuracy, sensitivity, and specificity.