Overview / Usage

Abnormal development of the spine will cause changes in the overall appearance of the back, such as the unequal height of high and low shoulders, scapula, and asymmetry of the profile of the left and right sides of the torso, which will seriously affect the physical appearance of people, and even damage their respiratory system function, motor function, and psychological state. Spine surgery is time-consuming, risky, and subjective. In the clinic, doctors classify the spine according to Lenke's classification rules. Doctors use tools such as Angle square and marker pen to measure Cobb Angle and perform Lenke typing according to the measurement Angle. Time-consuming and inaccurate. Our project uses a machine learning framework for spine segmentation tasks and automatic Cobb Angle measurement and Lenke automatic typing based on spine segmentation images. Specifically, the collected spinal images were fed into the target detection network to identify each pyramidal rectangular block of the spine, and the Cobb Angle was calculated by straight-line fitting through the center point of the rectangle.

This program can be used to help physicians identify scoliosis types online, making the identification process accurate and automated. We have previously developed an online scoliosis recognition website using CUDA +Tensorflow+ FAST-RCNN, but the recognition speed is relatively slow, so we can use oneAPI's heterogeneous programming technology to optimize the algorithm.