Approaches to building an automated diagnostic system for diseases of the lower extremities using pattern recognition methods
Artem E. Radin1, Julia A. Brodskaya1; 1Gagarin's State Technical University of Saratov, Saratov, Russia
Abstract
The paper proposes an algorithm for the automated diagnosis of diseases of the lower extremities (in particular, the foot) based on the analysis of X-ray images using deep learning methods, which reduces the time of primary screening. The developed method makes it possible to identify the relationships between remote areas of the image. A technique for preprocessing X-ray images is proposed, combining adaptive normalization of histograms taking into account projection distortions and a segmentation algorithm based on a modified U-Net architecture. This approach improves the accuracy of bone contour extraction by 18% compared to traditional methods when working with low-contrast images. A system of interpretation of the results has been developed, including both standard classification metrics and automatic calculation of clinically significant parameters. The algorithm provides an accuracy comparable to medical accuracy in measuring these indicators.
Speaker
Artem E. Radin
Gagarin's State Technical University of Saratov
Russia
Discussion
Ask question
