Automatic Thyroid Ultrasound Image Detection and Classification with Priori Knowledge

Abstract

Medical ultrasonic imaging technology is currently the preferred method to detect and diagnose benign and malignant thyroid nodules, which is widely used because of their low cost and non-invasive damage to patients. But automatic lesion detection and classification on thyroid ultrasound image is quite challenging due to the poor image quality. To solve the problem, based on popular Faster R-CNN network for natural image detection, a ResAt-Faster R-CNN model was proposed in the paper according to the characteristics of thyroid ultrasound image, the residual module and attention mechanism. The medical prior knowledges such as location and distribution information are further introduced to constrain the model to reduce the interference of surrounding tissues. The experimental results demonstrated that our proposed method was effective in the discrimination of thyroid nodules.