Fracture detection project for a Computer Vision course at Politecnico di Torino.

Leonardo Tanzi Fabien Cassassolles

The main idea is to use the combined action of two predictors.

• A set of classic Computer Vision filters are applied to the image
• A first predictor is used based on the Hough transform. It will detect the major lines in the image, and the angle they form with a reference. If several angles appear frequently, it might highlight a bone fracture.
• A second predictor is the presence of great angle values between the main lines of the X-ray image. They might result in a

• Refine parameters
• With a potential high CPU and GPU a Deep Learning approach could be interesting
  • CNN might be the optimal way to go as it is a standard approach to image recognition and could work for fracture classification
  • A Random Forest or a decision tree model could be a good way to start classifying diverse types of fractures