Research on Single-Object Tracking Algorithm for Thyroid Nodules Based on NanoTrack

With the rapid development of technology in the medical industry, hospitals have begun to introduce AI medical devices for auxiliary diagnosis. Among these, ultrasound AI medical imaging analysis technology has started to play a very important auxiliary role in the daily case diagnosis of doctors. This paper will use deep learning neural network technology to build a single-object tracking deep learning neural network, applied in the field of ultrasound imaging, to achieve single-object tracking of thyroid nodules. The algorithm will use three networks in a cascade. In the first frame of the video, the backbone network will extract deep features of the selected area of interest as the standard template. Starting from the second frame, the areas of interest in the image and the expanded regions of the image are sent into two different scale base networks for feature extraction. Finally, the features extracted by the two different base networks are fused and sent into a cascade network for classification and bounding box regression. This method locates the target by searching for the area that best matches the target template in each frame. It is a lightweight, high-speed tracking network. Experiments are deployed on the embedded device Jetson Nano, with a processing speed of up to 625 fps per second.