Abstract
The purpose of this study is to dynamically extract the screw-home movement (SHM) of the tibial surface from an ultrasonic device by semantic segmentation. SHM is a physiological phenomenon in which the tibia rotates extremely about 10 degrees with respect to the femur when the knee joint extends from the flexion position. Here, SHM is evaluated using a semantic segmentation model, U-Net as validation dataset. The validation shows that the final accuracy of the model is 0.55 for accuracy, which seems to be inappropriate for the SHM analysis. The extraction accuracy that is necessary to achieve depends on the field (e.g., medicine, sports) and on the purpose of semantic segmentation. Therefore, in this study we compare the pixel values of the bone center of the ultrasonic image and the center of the bone surface extracted by semantic segmentation to verify accuracy. If the center position in the ultrasonic image is output in the extracted image, the extraction accuracy has been verified. As a result of verification in 3 subjects, 56 of 60 observation points are output correctly. This result indicates the effectiveness of the semantic segmentation model used in this study to dynamically track the SHM of the tibial bone surface.
