Research Article
Prediction of Conversion to Alzheimer’s Disease Using 3D-DWT and PCA
@INPROCEEDINGS{10.1007/978-3-030-99197-5_16, author={Li Yew Aow Yong and Mohd Shafry Mohd Rahim and Chi Wee Tan}, title={Prediction of Conversion to Alzheimer’s Disease Using 3D-DWT and PCA}, proceedings={IoT Technologies for Health Care. 8th EAI International Conference, HealthyIoT 2021, Virtual Event, November 24-26, 2021, Proceedings}, proceedings_a={HEALTHYIOT}, year={2022}, month={3}, keywords={Alzheimer’s Disease Structural MRI DWT PCA}, doi={10.1007/978-3-030-99197-5_16} }- Li Yew Aow Yong
Mohd Shafry Mohd Rahim
Chi Wee Tan
Year: 2022
Prediction of Conversion to Alzheimer’s Disease Using 3D-DWT and PCA
HEALTHYIOT
Springer
DOI: 10.1007/978-3-030-99197-5_16
Abstract
Alzheimer’ Disease (AD) is the most common form of dementia worldwide. Structural Magnetic Resonance Imaging (sMRI) is the supportive tool for the diagnosis of this disease. Even, it can be used to predict the conversion of the disease from the mild cognitive impairment (MCI) to AD stage. Nevertheless, the 3D image produced by sMRI is high dimensional data, which raises the risk of overfitting in the classification model. For this reason, the combination of Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) was proposed as the feature extraction techniques to reduce the dimensional and extract significant features concurrently. The issues of DWT are the selection of level of decomposition and wavelet filter to decompose the image. In order to deal with these issues, a series of experiments were conducted to find the suitable parameters. By using 2D-DWT, spatial information of 3D data cannot be captured. The connection between the slices is neglected. Hence, 3D-DWT has been adopted instead of 2D-DWT in this paper. In the classification step, Support Vector Machine (SVM) was used as the classifier to predict the conversion of normal control (NC) and stable MCI (SMCI) to progressive MCI (PMCI) and AD for datasets collected up to 2 years before the progression. The dataset used in this paper was collected from Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. In the validation, the proposed method outperformed the other methods by attaining 79%, 79%, 82% and 82% in accuracy for the datasets collected at different time points, which were 1% to 4% higher than the model adopted 2D-DWT and PCA.
