Multiple training of neural networks for automatic spine segmentation
Open Access
- 21 June 2024
- journal article
- Published by V. N. Karazin Kharkiv National University in Bulletin of V.N. Karazin Kharkiv National University, series «Mathematical modeling. Information technology. Automated control systems»
- Vol. 62 (62) , 37-44
Abstract
Preventive and diagnostic studies of the presence of bone diseases require morphometric studies of X-ray images of the chest area. Nowadays, artificial intelligence methods are increasingly being used to solve such problems. The main difficulties of this task are related to the fact that X-ray images have quality limitations, for example, in terms of signal-to-noise ratio or contrast. For this reason, the application of standard methods of image recognition or automatic diagnosis becomes impossible. These difficulties have led to the fact that there is currently a fairly large number of works in this field, but the results of most of them are insufficient for practical use. Goal Investigate the possibility of using artificial intelligence in the segmentation of medical images for the purpose of automatic diagnosis of diseases of the human bone system. Research methods The ensemble method of X-ray image segmentation has been used in the study. The baseline of training data was created on the basis of X-ray images taken from open sources. The total number of images is 183. The initial data was modified according to the requirements necessary for model training. All images were converted to grayscale and resized to 256x256 pixels. Results Using this method in the two test cases resulted in an improvement in accuracy from 0.543 to 0.820 for the first snapshot and from 0.725 to 0.923 for the second snapshot. Conclusions We have proposed and investigated the application of the methodology of using an ensemble of reusable neural networks for automatic segmentation of a certain area of the spine, namely the Th8-Th11 spine region. The application of this method allowed obtaining more stable and accurate predictions for the desired spine regions, even for images with high noise levels.Keywords
This publication has 6 references indexed in Scilit:
- Automatic detection and segmentation of lumbar vertebrae from X-ray images for compression fracture evaluationComputer Methods and Programs in Biomedicine, 2021
- Vertebral corners detection on sagittal X-rays based on shape modelling, random forest classifiers and dedicated visual featuresComputer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2018
- U-Net: Convolutional Networks for Biomedical Image Segmentation2015
- Spinal Imaging and Image Analysis2015
- Vertebral morphometry: current methods and recent advancesEuropean Radiology, 2008
- Measures of the Amount of Ecologic Association Between SpeciesEcology, 1945