Automatic detection of Alzheimer's disease from EEG signals using low-complexity orthogonal wavelet filter banks. (March 2023)

• Record Type: Journal Article
• Authors: Puri, Digambar V.; Nalbalwar, Sanjay L.; Nandgaonkar, Anil B.; Gawande, Jayanand P.; Wagh, Abhay
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Alzheimer's disease -- Electroencephalogram -- Fractal dimension -- Orthogonal filter banks -- Support vector machine -- Wavelets; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104439
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Background: Alzheimer's disease (AD) is one of the most common neurodegenerative disorder. As the incidence of AD is rapidly increasing worldwide, detecting it at an early stage can prevent memory loss and cognitive dysfunctions in patients. Recently, Electroencephalogram (EEG) signals in AD cases show less synchronization and a slowing effect. The abrupt and transient behavior of EEG signals can be detected from specific frequency bands that are cortical rhythms of interest such as delta ( 0 − 4 Hz ), theta ( 4 − 8 Hz ), alpha ( 8 − 12 Hz ), beta1 ( 12 − 16 Hz ), beta2 ( 16 − 32 Hz ), and gamma ( 32 − 48 Hz ). Method: This paper proposes novel low-complexity orthogonal wavelet filter banks with vanishing moments (LCOWFBs-v) to decompose the AD and normal controlled (NC) EEG signals into subbands (SBs). A generalized design technique is suggested to reduce the computational complexity of original irrational wavelet filter banks (FBs). The two features, Higuchi's fractal dimension (HFD) and Katz's fractal dimension (KFD), were extracted from EEG SBs. The significance of these extracted features has been inspected using Kruskal–Wallis test. Results: The present study analyzed the EEG recordings of 23 subjects (AD-12 and NC-11) with the combination of LCOWFBs, HFD, and KFD. The proposed technique achieved a classification accuracy of 98.5% and 98.6% using the LCOWFBs-4 and LCOWFBs-6, respectively with a cubic -support vector machine classifier and 10-fold cross-validation technique. Conclusion: The proposed method with newly designed LCOWFBs is efficient compared with the well-known FBs and existing techniques for detecting AD. Graphical abstract: Highlights: Novel orthogonal wavelet filters are designed to reduce computational complexity. The newly designed wavelet filter banks are rational. This study utilizes an efficient feature set through a statistical KW test. We proposed wavelet-based HFD and KFD features with cubic-SVM for detection of AD. The proposed model achieved highest accuracy using LOWFBs-6, HFD, and KFD features. The proposed model outperforms the existing techniques of AD detection.