- Title
- Accurate multiple sclerosis detection and prediction using advanced image processing and deep learning
- Creator
- Afzal, Hafiz Muhammad Rehan
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2021
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Multiple Sclerosis (MS) is a chronic neurological disease of the central nervous system (CNS). Early diagnosis of MS is highly desirable since treatments are more effective in preventing MS-related disability if commenced in the early stages. This thesis starts with an overview of MS and artificial intelligence (AI). Then the diagnostic procedure for MS is described in detail. Diagnostic criteria (the McDonald Criteria) and guidelines for imaging (Barkhof’s and MAGNIMS) are discussed. Understanding diagnostic criteria is essential in order to develop automated algorithms for MS diagnosis and monitoring of disease activity. After that, a literature review is presented on recent progress in state-of-the-art AI algorithms for MS. I then describe the three algorithms I have developed as part of this research program for segmentation, detection, and prediction of further MS events. The aims of my research were to, firstly, segment MS lesions and, secondly, predict the occurrence of a second MS-related clinical event, which manifests as conversion of clinically isolated syndrome (CIS) to clinically definite MS (CDMS). Both tasks – segmentation and prediction – are based on the analysis of MRI scans using a branch of AI known as ‘deep learning’. Deep learning detects high level features from data which makes it a step ahead of machine learning. The diagnosis of MS is predominantly based on the detection of lesions on magnetic resonance imaging (MRI) which also provides ongoing essential information about progression and disease activity. Manual detection of lesions is time consuming and lacks accuracy. Especially cortical lesions are difficult to detect manually. It is particularly difficult to detect volumetric changes over time. The first algorithm was developed to detect MS lesions. A novel and fully automated convolutional neural network (CNN) approach was developed to segment lesions. The developed system consists of two 2D patch wise CNNs which can segment lesions accurately and robustly. The first CNN network is implemented to segment lesions accurately, and the second network is implemented to reduce false positives to increase efficiency. The system consists of two parallel convolutional pathways, where one pathway is concatenated to the second and, at the end, the fully connected layer is replaced with a CNN. Three routine MRI sequences T1-w, T2-w and FLAIR were used as input to the CNN. FLAIR was used for segmentation because most lesions on MRI appear as bright regions and T1-w & T2-w were used to reduce MRI artifacts. The system was evaluated on two challenge datasets that are publicly available from Medical Image Computing and Computer-Assisted Intervention (MICCAI) and International Symposium on Biomedical Imaging (ISBI). Quantitative and qualitative evaluation has been performed with various metrics including false positive rate (FPR), true positive rate (TPR) and dice similarities, and were compared to current state-of-the-art methods. The method shows consistent higher precision and sensitivity than other methods. The method can accurately and robustly segment MS lesions from images produced by different MRI scanners, with a precision of 90%. For the next part of my research, I focused on prediction. For the prediction of conversion from CIS to CDMS, two algorithms were developed. The first algorithm was derived from a small dataset and had a simple architecture. The algorithm used improved CNNs that used LeNet architecture coded in Python and the Keras library. The algorithm consisted of convolutional layers, which learned the patterns of input images by using convolutional filters. The algorithm predicted conversion from CIS to CDMS with 83.3% and 100% accuracy in two experiments (average of 91.6%). The results are described in detail in Chapter 5. With more scans collected, a third algorithm was developed which is more complex and robust. The basic architecture of this algorithm is that of the VGG16 CNN model, but altered such that it can handle MRI Digital Imaging and Communications in Medicine (DICOM images.) A dataset comprised of scans acquired using two different scanners was used for the purpose of verification of the algorithm. A group of 69 patients had volumetric MRI scans taken at onset of the disease and then again two years later using one of the two scanners. In total, this yielded 11,40 images which were then used for training, validation and testing of the algorithm. Initially, these raw images were taken through 4 steps of pre-processing. In order to boost the efficiency of the process, the algorithm was pretrained using the publicly available Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset used to classify Alzheimer’s Disease. Finally, the pre-processed dataset was used to train and test the algorithm. Clinical evaluation conducted two years after the first time point revealed that 35 of the 69 patients had converted to CDMS, while the remaining 34 had not. Results of testing showed that the algorithm was able to predict the clinical results with an accuracy of 89.3%, with an area under the curve (AUC) of 91%. The robustness of the algorithm was evident in its consistently high accuracy when analysing datasets derived from the two different scanners. These three algorithms (segmentation, detection and prediction from CIS to CDMS), make a full system which will help neurologists to segment, detect and predict MS disease automatically, without any human interaction. These algorithms can be used to prioritise high efficacy therapy to patients converting to MS early as compared to other patients. Further algorithms might also predict the level of disability at later stages of the disease.
- Subject
- artificial intelligence; multiple sclerosis; McDonald criteria; MS disease activity; convolutional neural networks; CNN; prediction of disease; machine learning; deep learning; automatic disease prediction; neurological disease; lesions segmentation; brain segmentation; diagnosis of multiple sclerosis
- Identifier
- http://hdl.handle.net/1959.13/1473752
- Identifier
- uon:49100
- Rights
- Copyright 2021 Hafiz Muhammad Rehan Afzal
- Language
- eng
- Full Text
- Hits: 2016
- Visitors: 2094
- Downloads: 146
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details Download | ATTACHMENT01 | Abstract | 330 KB | Adobe Acrobat PDF | View Details Download | ||
View Details Download | ATTACHMENT02 | Thesis | 1 MB | Adobe Acrobat PDF | View Details Download |