Computer-aided diagnosis system for bone fracture detection using machine learning algorithms

Naderian, A, 2022. Computer-aided diagnosis system for bone fracture detection using machine learning algorithms. PhD, Nottingham Trent University.

[thumbnail of Amirkhashayar Naderian 2024.pdf]
Preview
Text
Amirkhashayar Naderian 2024.pdf - Published version

Download (1MB) | Preview

Abstract

Diagnostic imaging technology has revolutionized the healthcare industry by allowing more accurate and earlier diagnosis of diseases. This technology reduces the need for invasive procedures such as surgery and enhances the quality of patient care. Several machine learning algorithms like SVM, K-mean clustering and UNET have been demonstrated to be capable of solving classification, detection, and segmentation problems in medical imaging, as well as being used for super-resolution techniques. The purpose of this thesis is to examine machine learning and image processing methods for four key challenges in medical image analysis.

The first one is the segmentation of medical images. The second challenge involves implementing super-resolution techniques for medical images. Third, using image processing methods in order to diagnose the abnormalities. The fourth contribution is to enrich the image information by mapping of medical images between different modalities using deep neural models. In this research, all contributions aim at developing an end-to-end model that can detect fractures automatically or be used as a clinical assistant to reduce errors. As the first contribution, the thesis presents a multi-stage novel approach for bone segmentation in X-ray images using faster region-based convolutional neural network (R-CNN) and distance regularized level set evolution (DRLSE) algorithms. A hybrid model utilizing deep neural network (DNN) and image processing techniques are proposed to segment the bones in two stages. Our model is more robust to the changes in X-ray images, as well as applicable to bones that are misplaced. Additionally, we have used transfer learning to reduce the amount of time and effort required to collect and label the data. As the second contribution, DNN models are used to enhance the resolution of medical images. CNN and generative adversarial network have been used as super-resolution techniques to achieve high-resolution medical images. The analysis includes subjective and objective evaluations of different models on regions with or without fractures to compare them with our model. The third contribution involves applying different image analysis methods to X-ray images in order to detect fractures with the minimum amount of human intervention. By using entropy and intensity, we have also attempted to identify regions of interest that have a higher probability of having fractures. We also evaluate the effect of super-resolution technique on the saliency map with and without fractures. Lastly, we present image-to-image mapping by using variational autoencoders and generative adversarial networks to reduce the cost of diagnosis and medical images retrievals. We have attempted to map X-ray images to MRIs in this section in order to fuse the high diagnostic information existing in MRIs, for enhancing the matched X-ray images.

Item Type: Thesis
Creators: Naderian, A.
Contributors:
Name
Role
NTU ID
ORCID
Sanei, S.
Thesis supervisor
CMP3SANEIS
Kanjo, E.
Thesis supervisor
CMP3KANJOE
Date: November 2022
Rights: Amirkhashayar Naderian: Computer-aided diagnosis system for bone fracture detection using machine learning algorithms, A thesis submitted to Nottingham Trent University in candidature of the Degree of Doctor of Philosophy, © November 2022
Divisions: Schools > School of Science and Technology
Record created by: Melissa Cornwell
Date Added: 10 Apr 2024 13:48
Last Modified: 10 Apr 2024 14:36
URI: https://irep.ntu.ac.uk/id/eprint/51229

Actions (login required)

Edit View Edit View

Statistics

Views

Views per month over past year

Downloads

Downloads per month over past year