AI Improves Junior Radiologists’ Productivity

Background: Ultrasound detection and evaluation of thyroid nodules is a common imaging examination, but its results can be very operator dependent, causing significant variation in interpretations and clinical guidance. Artificial intelligence (AI) algorithms are showing promise in analyzing image data and providing valid management conclusions. Objective: To determine if AI algorithms developed for the evaluation of thyroid nodules by ultrasound could provide accurate interpretations that could add value to patient clinical management. Methods: Data on patients who underwent thyroid ultrasound were obtained from 3 Chinese medical centers. The thyroid ultrasounds were acquired following the standard thyroid ultrasound protocol. Five radiologists chose images that had well-documented transverse and longitudinal planes of each nodule. These de-identified images were sent to the AI program, which used AI software based on the EfficientNet architecture with a proprietary deep learning fr more...