88 ^th Annual Meeting of the American Thyroid Association

Abstract

Thyroid ultrasound screening has been involved in routine physical examination at hospitals. However, due to the uneven development of medical resources, the diagnostic accuracy for thyroid nodules varies greatly. The aim of this study was to explore a novel artificial intelligence-assisted ultrasound diagnostic system to improve the diagnosis efficiency and accuracy for thyroid nodules. The artificial intelligence-assisted ultrasound diagnostic system was developed based on computer deep learning, training and multi-frame dynamic image recognition techniques. In order to validate the accuracy of the system, 500 pathologically confirmed thyroid nodules were selected, including 208 benign and 292 malignant. The images of all nodules, acquired from 10 different types of ultrasound equipment, were dynamically stored in the form of the consecutively longitudinal and transverse sections. Then the AI-assisted diagnostic system automatically recognized and analyzed the features of the images, and gave recommendations for diagnosis. The diagnostic accuracy of the system was compared with that of junior and senior physicians. The results showed that the sensitivity of the AI-assisted diagnostic system, junior physicians, senior physicians, combo of junior physicians and the AI-assisted system was 82.9%, 52.6%, 87.4% and 98.6%, respectively; and the specificity was 70.2%, 95.7%, 83.2% and 83.7%, respectively. The diagnostic accuracy of the AI-assisted system alone was higher than that of junior physicians (77.6% vs. 70.5%), however, the accuracy can reach 92.4% when the junior physicians was assisted by the AI-assisted system, which was higher than that of the senior physicians (85.6%). At the same time, the diagnostic tests on thyroid nodules of different sizes (<1cm, 1-2cm, and >2cm) were performed using the system. The results showed that there was no significant difference in the diagnostic accuracy among the three groups (P < 0.05). The AI-assisted ultrasound diagnostic system can effectively assist ultrasound physicians in the diagnosis of benign and malignant thyroid nodules, reduce unnecessary puncture and over-treatment, and help the junior physicians to improve the diagnosis efficiency and accuracy.