A Single-Shot Region-Adaptive Network for Myotendinous Junction Segmentation in Muscular Ultrasound Images.

Researchers

Journal

Modalities

Models

Abstract

Tracking the myotendinous junction (MTJ) in consecutive ultrasound images is crucial for assessing the mechanics and pathological conditions of the muscle-tendon unit. However, poor image quality and boundary ambiguity conspire towards a lack of reliable and efficient identification of MTJ, restricting its application in motion analysis. In recent years, with the rapid development of deep learning, the region-based convolution neural network (RCNN) has shown great potential in the field of simultaneous objection detection and instance segmentation in medical images. This paper proposes a regionadaptive network, called RAN, to adaptively localize MTJ region and segment it in a single shot. Our model learns salient information of MTJ with a composite architecture, in which a region-based multi-task learning network explores the region containing MTJ, while a parallel end-to-end U-shape path extracts the MTJ structure from the adaptively selected region for combating data imbalance and boundary ambiguity. By demonstrating on ultrasound images of the gastrocnemius, we showed that the RAN achieves superior segmentation performance compared to the state-of-the-art Mask RCNN method with average Dice scores of 80.1%. Our method is promising in advancing muscle and tendon function examinations with ultrasound imaging.