Stroke Lesion Segmentation And Classification For Diffusion-Weighted Imaging
Magnetic Resonance Imaging (MRI) plays an important role in the diagnosis of brain disorders. Stroke is one of the major categories of brain disorders. Recent studies support the notion of stroke as the “time is brain” due to the fact that if the treatment is done within six hours of suffering a str...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/25380/1/Stroke%20Lesion%20Segmentation%20And%20Classification%20For%20Diffusion-Weighted%20Imaging.pdf http://eprints.utem.edu.my/id/eprint/25380/2/Stroke%20Lesion%20Segmentation%20And%20Classification%20For%20Diffusion-Weighted%20Imaging.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Magnetic Resonance Imaging (MRI) plays an important role in the diagnosis of brain disorders. Stroke is one of the major categories of brain disorders. Recent studies support the notion of stroke as the “time is brain” due to the fact that if the treatment is done within six hours of suffering a stroke, the patient's life can be saved and the outcome can be improved. Conventionally, the diagnosis of brain stroke is performed manually by professional neuroradiologists during a highly subjective and time- consuming process. Therefore, this study proposes a technique for automatic detection, segmentation and classification of brain stroke lesion from MRI images. The types of stroke lesion are acute hemorrhage stroke, acute ischemic stroke, chronic ischemic stroke and sub-acute ischemic stroke. Diffusion weighted imaging (DWI) sequences from the MRI is chosen for the analysis using machine learning and deep learning techniques. The machine learning technique consists of four stages which are pre-processing, segmentation, features extraction and classification. For segmentation, adaptive thresholding, gray level co-occurrence matrix (GLCM), marker-controlled watershed, fuzzy c-Means (FCM) and kMeans are proposed to segment the stroke region. Statistical features are calculated and fed into several classification techniques, which are the linear discriminant analysis (LDA), support vector machine (SVM), weighted k- Nearest Neighbor (k-NN) and bagged tree classifier. Deep learning using regional convolutional neural network (R-CNN) technique is also proposed in the analysis. The technique consists of four stages which are input image, Region Proposal Network (RPN), Convolutional Neural Network (CNN) features computation and classification. The segmentation performances are evaluated using Jaccard indices, Dice Coefficient, false positive and false negative rates. For classification, the performances are evaluated using accuracy, sensitivity and specificity. Segmentation results demonstrated that k-Means offered the best performance for stroke lesion segmentation while sub-acute ischemic stroke gave the highest rate with 0.85 Dice index. Results demonstrated that support vector machine (SVM) offered the best performance for stroke lesion classification with accuracy 98.5% and average training time is 1.8 second. In conclusion, the proposed stroke classification technique has the potential to diagnose and classify brain stroke lesions. |
|---|