A robust swarm intelligence-based feature selection model for neuro-fuzzy recognition of mild cognitive impairment from resting-state fMRI

Abstract Individuals diagnosed with mild cognitive impairment (MCI) are at a high risk of transition to Alzheimer's disease (AD), but a diagnosis of MCI is challenging. Resting-state functional magnetic resonance imaging (rs-fMRI) is a promising tool for identifying patients with MCI, but an accurate and robust analysis method is needed to extract discriminative rs-fMRI features for classification between MCI patients and healthy people. In this paper, a new rs-fMRI data analysis approach based on Chaotic Binary Grey Wolf Optimization (CBGWO) and Adaptive Neuro-Fuzzy Inference System (ANFIS), namely (CBGWO-ANFIS), is presented to distinguish MCI patients based on rs-fMRI. CBGWO is a new feature selection model that attempts to reduce the number of features without loss of significant information for classification, and it uses the naive Bayes criterion as a part of the objective function. Based on the chaos theory, the important parameters of GWO are estimated and tuned by using ten different chaos sequence maps. Subsequently, ANFIS is used to classify MCI patients and healthy people based on the subset of features retained by CBGWO. Experiments were carried out on 62 MCI patients and 65 normal controls (NC). Fractional amplitude of low frequency fluctuation (f-ALFF) was extracted from rs-fMRI as features. The results indicate that the proposed CBGWO-ANFIS approach with the Chebyshev chaos map shows a higher accuracy (around 86%), higher convergence speed, and shorter execution time than other chaos maps. Further, the proposed approach outperforms the conventional machine learning techniques and the recent meta-heuristic optimization algorithms. This study indicates that the proposed CBGWO-ANFIS approach on rs-fMRI could be a potential tool for early diagnosis of MCI.