Papillary thyroid cancer (PTC) accounts for approximately 90% of all thyroid malignancies, with ultrasound playing a critical role in diagnosis. Although artificial intelligence (AI)-assisted systems have shown promise in improving diagnostic accuracy, widespread adoption is hindered by concerns regarding cost and data security. To address these issues, we developed a smartphone-compatible AI diagnostic system targeting ultrasound images of PTC. We used ultrasound images from 150 PTC and 200 benign thyroid lesions to train an AI server employing an object detection model for lesion identification. A smartphone application was developed to communicate with the AI server and display diagnostic predictions. For performance evaluation, we used 15 PTC and 15 benign images extracted from the training dataset. Sensitivity, specificity, and overall accuracy were calculated for both server-based and smartphone-based predictions. The AI server achieved a sensitivity of 100%, specificity of 96.7%, and overall accuracy of 97.8%. The smartphone-based system also showed high performance, with a sensitivity of 100%, specificity of 90.0%, and overall accuracy of 93.3%. One instance of misclassification occurred, where tracheal cartilage was incorrectly identified as a PTC. The developed AI system enables highly accurate diagnosis of PTC from ultrasound images. The smartphone interface retains robust diagnostic performance, offering a low-cost, secure, and accessible tool for clinical use, particularly in resource-limited settings.
Takahashi et al. (Wed,) studied this question.