Research on Multi-Object Tracking Algorithm for Thyroid Nodules Based on ByteTrack

The field of medical technology has begun to use artificial intelligence technology for auxiliary diagnosis. This article uses multi-target tracking technology to automatically track and locate thyroid nodules. In the application scenario of ultrasound imaging, this article uses a tracking-by-detection method. The multi-object tracking model ByteTrack performs tracking by associating each detection box instead of only associating high-scoring detection boxes. For low-scoring detection boxes, their similarity to trajectories is exploited to recover real objects and filter out background detections. Using the similarity between the detection frame and the tracking trajectory, while retaining the high-scoring detection results, the background is removed from the low-scoring detection results to discover the real objects (difficult samples such as occlusion and blurring), thereby reducing missed detections and maintain track connectivity. This method uses Yolov7 as the thyroid nodule target detector and ByteTrack as the multi-target tracker of thyroid nodules to ensure that thyroid nodules can be accurately tracked in real-time scenarios, complete the end-to-end input and output of the algorithm, and deploy the algorithm on edge devices, the tracker can process up to 1256 fps.