Research And Application Of Muscle Strain Recognition Algorithm Based On Deep Learning

The recognition and analysis of medical images of b-mode muscular ultrasound is very important for doctors' clinical diagnosis.In the past,doctors used to recognize and analyze patients' medical images of b-mode muscular ultrasound based on their clinical experience,and then make a diagnosis.This method is influenced by experience and subjective thinking,and it is even more difficult and challenging for inexperienced doctors.According to the characteristics of b-ultrasound medical images of muscle and combined with doctors' clinical experience,this paper proposes a deep learning-based method for the identification of muscle strain area in B-ultrasound medical images,which is mainly divided into two steps: detection and filtering.Detection refers to the use of Yolov3 algorithm to preliminarily identify muscle strain in a single b-ultrasound image and obtain candidate areas of strain.Filter refers to the use of sliding window and considerations of HOG feature to find the candidate region corresponding to the reference area,and then using the siamese neural network,combined with the strain candidate area corresponding to the reference area,to further filter of strain candidate area,excluding low confidence candidate strain area and artifacts such as interference,so that the end result is more accurate.Finally,the test was carried out on the test set of 32 medical b-ultrasound images of the patients.The predicted positions on each image were ranked according to their confidence from large to small,and there were strain loss areas in the three predicted positions with the highest confidence of the medical b-ultrasound images of the muscles,accounting for 93.75%.On average,there will be 1 muscle strain region in every 3 regions predicted by the algorithm.The method proposed in this paper is an exploration of the identification of muscle strain area in b-ultrasonic medical images,and it promotes the improvement of doctors' diagnostic level by giving play to the auxiliary analysis function of technology.