An Automatic Knee Osteoarthritis Diagnosis Method Based on Deep Learning: Data from the Osteoarthritis Initiative

Figure 2

Overview of the proposed method. The OA severity assessment is divided into four steps. (1) An object detection CNN extracts the bonding boxes of the knees from the X-ray images. The confidence scores of the detection are used to filter the low-quality detection. (2) A CNN backbone extracts the spatial feature maps for the cropped knee images. (3) The spatial feature maps are flatted and recomposed as a sequence. (4) The visual transformer exploits the correlations between different local regions for the final classification task.