Cover Image

Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability

Many past studies had used multiple MRI scans and protocols to automate the prediction of MS patients’ disability. They focused on using non-raw MRI data including clinical, radiological, and general patient information with different study durations. Furthermore, they were using manual and semi-aut...

Full description

Saved in:
Bibliographic Details
Main Author: M. Muslim, Ali
Format: Thesis
Language:English
Published: 2022
Subjects:
Image data mining
Biomedical engineering
Online Access:http://psasir.upm.edu.my/id/eprint/114906/1/114906.pdf
http://psasir.upm.edu.my/id/eprint/114906/
http://ethesis.upm.edu.my/id/eprint/18206
Tags: Add Tag
No Tags, Be the first to tag this record!
id my.upm.eprints.114906
record_format eprints
spelling my.upm.eprints.1149062025-02-25T02:32:51Z http://psasir.upm.edu.my/id/eprint/114906/ Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability M. Muslim, Ali Many past studies had used multiple MRI scans and protocols to automate the prediction of MS patients’ disability. They focused on using non-raw MRI data including clinical, radiological, and general patient information with different study durations. Furthermore, they were using manual and semi-automated features extraction. Unlike previous studies, this study aims to predict MS patients’ disability by using automated feature extraction, single MRI scan, and single MRI protocol, without patient follow up. Since each part of the brain controls a specific human body function, the location of brain abnormalities in which lobes would help to identify the type of dysfunction, and at which part of the human body. Different brain abnormality’s location may result in different values of MS patient disability scores. Thus, segmenting the brain abnormalities that have a high correlation to the patient’s disability and classifying them according to their locations would be significant for disability prediction. This study uses data extracted from 65 MS patients who were from multiple centers in Iraq and Saudi Arabia. The Dynamic Image Thresholding (DIT) method was proposed to segment areas of brain abnormalities on brain MRI. This is followed by an estimation method to segments the brain lobes and brain periventricular region segmentation (BLBPRS). The performance of DIT and BLBPRS methods were evaluated by two experts, radiologists, for each method with an overall performance evaluation of 80% and 79% respectively. A large-scale statistical, volumetric, texture, location, radiological, clinical and ratio-based features were extracted using clinical, radiological, general patient information, and raw-imaging data. From the large-scale features, a correlation analysis is performed to select the highly correlated features used for predicting patients’ disability. This was based on machine learning and regression algorithms at the first phase. The proposed methodology is divided into two phases. The first phase aims to investigate the best types of required data, features and algorithms to be used in the final proposed methodology to predict exact EDSS, and different ranges of EDSS. A 5-fold cross-validation has been used to evaluate the performance. In the first phase, all dataset is combined and weak performance was found. In the second phase, the dataset was divided into four groups according to the MRI-Tesla and the condition of a lesion in the spinal cord or not. The division of dataset into four groups produced good performance in EDSS prediction and classification. The best machine learning performance, after the grouping, came from SVM, with an average accuracy, sensitivity, and specificity of 82%, 77%, and 79%, respectively. The best performance from the linear regression had an average RMSE of 0.6 for EDSS step of 2. These results showed the possibility of using fully automated feature extraction, single MRI scan, and single MRI protocols without patient follow-up to predict MS patients’ disability. 2022-09 Thesis NonPeerReviewed text en http://psasir.upm.edu.my/id/eprint/114906/1/114906.pdf M. Muslim, Ali (2022) Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability. Doctoral thesis, Universiti Putra Malaysia. http://ethesis.upm.edu.my/id/eprint/18206 Image data mining Biomedical engineering
institution Universiti Putra Malaysia
building UPM Library
collection Institutional Repository
continent Asia
country Malaysia
content_provider Universiti Putra Malaysia
content_source UPM Institutional Repository
url_provider http://psasir.upm.edu.my/
language English
topic Image data mining
Biomedical engineering
spellingShingle Image data mining
Biomedical engineering
M. Muslim, Ali
Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
description Many past studies had used multiple MRI scans and protocols to automate the prediction of MS patients’ disability. They focused on using non-raw MRI data including clinical, radiological, and general patient information with different study durations. Furthermore, they were using manual and semi-automated features extraction. Unlike previous studies, this study aims to predict MS patients’ disability by using automated feature extraction, single MRI scan, and single MRI protocol, without patient follow up. Since each part of the brain controls a specific human body function, the location of brain abnormalities in which lobes would help to identify the type of dysfunction, and at which part of the human body. Different brain abnormality’s location may result in different values of MS patient disability scores. Thus, segmenting the brain abnormalities that have a high correlation to the patient’s disability and classifying them according to their locations would be significant for disability prediction. This study uses data extracted from 65 MS patients who were from multiple centers in Iraq and Saudi Arabia. The Dynamic Image Thresholding (DIT) method was proposed to segment areas of brain abnormalities on brain MRI. This is followed by an estimation method to segments the brain lobes and brain periventricular region segmentation (BLBPRS). The performance of DIT and BLBPRS methods were evaluated by two experts, radiologists, for each method with an overall performance evaluation of 80% and 79% respectively. A large-scale statistical, volumetric, texture, location, radiological, clinical and ratio-based features were extracted using clinical, radiological, general patient information, and raw-imaging data. From the large-scale features, a correlation analysis is performed to select the highly correlated features used for predicting patients’ disability. This was based on machine learning and regression algorithms at the first phase. The proposed methodology is divided into two phases. The first phase aims to investigate the best types of required data, features and algorithms to be used in the final proposed methodology to predict exact EDSS, and different ranges of EDSS. A 5-fold cross-validation has been used to evaluate the performance. In the first phase, all dataset is combined and weak performance was found. In the second phase, the dataset was divided into four groups according to the MRI-Tesla and the condition of a lesion in the spinal cord or not. The division of dataset into four groups produced good performance in EDSS prediction and classification. The best machine learning performance, after the grouping, came from SVM, with an average accuracy, sensitivity, and specificity of 82%, 77%, and 79%, respectively. The best performance from the linear regression had an average RMSE of 0.6 for EDSS step of 2. These results showed the possibility of using fully automated feature extraction, single MRI scan, and single MRI protocols without patient follow-up to predict MS patients’ disability.
format Thesis
author M. Muslim, Ali
author_facet M. Muslim, Ali
author_sort M. Muslim, Ali
title Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
title_short Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
title_full Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
title_fullStr Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
title_full_unstemmed Automated feature extraction on brain MRI images for predicting multiple sclerosis patient disability
title_sort automated feature extraction on brain mri images for predicting multiple sclerosis patient disability
publishDate 2022
url http://psasir.upm.edu.my/id/eprint/114906/1/114906.pdf
http://psasir.upm.edu.my/id/eprint/114906/
http://ethesis.upm.edu.my/id/eprint/18206
_version_ 1825162396038070272
score 13.239859

Similar Items