Artificial intelligence-driven dynamic ultrasound system and method for detecting subacromial impingement

An artificial intelligence-driven dynamic ultrasound method for detecting subacromial impingement, comprising the following steps of:

• (a) generating a specific number of images from a dynamic video during a specific period at a frequency of four frames per second, and inputting the images into a trained Faster Region with Convolution Neural Network (Faster R-CNN) module, wherein a ResNet50 architecture in the Faster R-CNN module directly extracts a feature of each of the images with a layer frame and generates at least one bounding box in each of the images;
• (b) receiving the bounding box by a region proposal network (RPN) architecture, and correcting the bounding box to filter out at least one candidate box with a bone of a shoulder;
• (c) receiving the candidate box and performing a region of interest pooling (ROI pooling) by a first fully connected layer in the Faster R-CNN module, wherein a regional length of the candidate frame is fixed, followed by classifying;
• and outputting the candidate frame with the bone of the shoulder;
• (d) filtering out the candidate box with a highest probability of appearing of the bone of the shoulder according to the bone of the shoulder in the candidate box, recording and locating an exact position of the candidate frame to convert the exact position into a coordinate position by the first fully connected layer, so as to draw an action trajectory;
• and
• e) converting the action trajectory into a frequency domain, inputting the frequency domain into a trained 1D-convolutional neural network (1D-CNN) module, and outputting a probability of developing subacromial impingement syndrome by the 1D-CNN module.