Segmentační model pro predikci hojení kostí z RTG obrazů

Abstract

This thesis focuses on creating a model for predicting bone healing from X-ray images. The theoretical part of the work explains the basic principles of X-ray images and how bone fractures can be analyzed from them. It also describes different methods for dividing the image into individual parts. In the practical part of the work, two datasets containing X-ray images of broken bones were created. Then, a segmentation model was developed, which can distinguish fractures from the background based on adaptive thresholding and active contours. The results of the segmentation were analyzed, and the model was tested using the Sørensen-Dice coefficient. The first dataset was used to compare the brightness of the fracture with that of a healthy bone, while the second dataset was used to compare the properties of the fracture over time, such as perimeter, area, and brightness. The results showed that the segmentation model achieved an average accuracy of 0.8861, where 1 was the best possible result. In the analysis of the dynamics of the healing processes, it was found that the brightness intensity of the fracture increases over time, while the perimeter and area of the fracture decrease.