Classification of Alzheimer's Disease using Low Frequency Fluctuation of rs-fMRI Signals

Abstract

The resting-state functional magnetic resonance imaging (rs-fMRI) is a non-invasive neuroimaging modality to measure brain activity and helps in the diagnosis of various brain-related disorders. Given the 1/f power spectrum characteristic of brain dynamics, where the energy value is higher at a low frequency than high frequency, it is established that low-frequency oscillations (LFO) provide a better representation of the spontaneous neuronal activity of the brain. In this research, a combination of the amplitude of low-frequency fluctuation (ALFF) and fractional ALFF (fALFF) from the resting-state blood oxygen level-dependent (BOLD) signal in the classic band i.e., 0.01-0.1 Hz is used for the classification of Alzheimer's disease (AD) from normal controls (NC). A total of 60 subjects participated in this study consisting of 30 AD patients and 30 NC from Alzheimer's disease neuroimaging initiative (ADNI). The feature selection is performed using minimum-redundancy maximum-relevance (mRMR) and ReliefF algorithm due to the large dimension of rs-fMRI data to be fed to the machine learning (ML) classifier. The proposed AD classification method employing the fusion of ALFF and fALFF obtained the highest classification accuracy of 96.36%, indicating the good potential of the proposed method for the diagnosis of AD, as well as other neurological conditions.