Enhanced alzheimer’s disease classification scheme using 3d features
Alzheimer’s disease (AD) is a neurodegenerative brain illness that leads to death due to complications. Many studies on AD classification with Magnetic Resonance Imaging (MRI) images were conducted to act as a computer-aided diagnosis. Feature extraction and feature selection were performed to reduc...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/101528/1/AowYongLiYewPSC2021.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Alzheimer’s disease (AD) is a neurodegenerative brain illness that leads to death due to complications. Many studies on AD classification with Magnetic Resonance Imaging (MRI) images were conducted to act as a computer-aided diagnosis. Feature extraction and feature selection were performed to reduce the number of features and extract significant features concurrently. However, the classification of stable mild cognitive impairment (SMCI) and progressive mild cognitive impairment (PMCI) is far from satisfactory due to the high similarity between the groups. Therefore, this research aimed to enhance the AD classification scheme to solve the problem. The proposed method has included shape enhancement before feature extraction to maximize the difference between healthy patients (normal control (NC)+SMCI) and sick patients (PMCI+AD). The sick patient has a thinner brain boundary compared to a healthy patient. Therefore, a 3D opening morphological operation was proposed to eliminate the thinner boundary and restore the thicker boundary. After that, the proposed 3-level 3D Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) were combined for feature extraction. Using the Haar filter, 3-level 3D-DWT extracted 3D significant features to improve the classification result. PCA further reduced the number of features by projecting the training set and test set to lower-dimensional space. The number of features was greatly reduced from 2,122,945 to 159. Feature selection was removed from the proposed scheme after realizing the process would eliminate important features to segregate the classification groups. Linear Support Vector Machine (SVM) was employed to perform binary classification. The proposed scheme achieved higher mean accuracy compared to the previous method, which was from 79% to 80%, from 81% to 84%, from 80% to 84 % on the datasets collected at time points of 24 months, 18 months before stable diagnosis and at the stable diagnosis time point, respectively. |
|---|