Model-based tendon segmentation from ultrasound images

In orthopedics, trigger finger is one of the popular occupational hazards in recent years. Ultrasound images are usually used for diagnosing the severity of trigger finger clinically. Finger ultrasound image has two important characteristics: the shape of tendon is close to an ellipse, and the tendon boundaries vary significantly in image appearance. The traditional segmentation methods usually cannot segment the tendon well. In this study, we develop an ultrasound image detection and estimation system that can assist clinician to locate and evaluate the area of tendon and synovial sheath automatically. An adaptive texture-based active shape model (ATASM) method is proposed to overcome the complex segmentation problems with the proposed shape model by minimizing the objective function based on gradient and texture information. Considering the segmentation may have many local solutions due to various image qualities, the genetic algorithm (GA) is adopted to search for the best shape parameters. In the experiments, the results of tendon segmentation are found with small segmentation errors and similar to the contour drawn by trained users.