[Paper Review] Automatic diagnosis of schizophrenia and attention deficit hyperactivity disorder in rs-fMRI modality using convolutional autoencoder model and interval type-2 fuzzy regression

Afshin Shoeibi, Navid Ghassemi+8 more|arXiv (Cornell University)|May 31, 2022

EEG and Brain-Computer Interfaces129 references70 citations

TL;DR

This paper proposes a novel deep learning and fuzzy logic framework for automated diagnosis of schizophrenia (SZ) and attention deficit hyperactivity disorder (ADHD) using resting-state fMRI (rs-fMRI) data. It employs a custom 2D convolutional autoencoder (CNN-AE) for feature extraction and introduces an interval type-2 fuzzy regression (IT2FR) classifier optimized via gray wolf optimization (GWO), achieving 72.71% accuracy in distinguishing healthy controls, SZ, and ADHD patients.

ABSTRACT

Nowadays, many people worldwide suffer from brain disorders, and their health is in danger. So far, numerous methods have been proposed for the diagnosis of Schizophrenia (SZ) and attention deficit hyperactivity disorder (ADHD), among which functional magnetic resonance imaging (fMRI) modalities are known as a popular method among physicians. This paper presents an SZ and ADHD intelligent detection method of resting-state fMRI (rs-fMRI) modality using a new deep learning method. The University of California Los Angeles dataset, which contains the rs-fMRI modalities of SZ and ADHD patients, has been used for experiments. The FMRIB software library toolbox first performed preprocessing on rs-fMRI data. Then, a convolutional Autoencoder model with the proposed number of layers is used to extract features from rs-fMRI data. In the classification step, a new fuzzy method called interval type-2 fuzzy regression (IT2FR) is introduced and then optimized by genetic algorithm, particle swarm optimization, and gray wolf optimization (GWO) techniques. Also, the results of IT2FR methods are compared with multilayer perceptron, k-nearest neighbors, support vector machine, random forest, and decision tree, and adaptive neuro-fuzzy inference system methods. The experiment results show that the IT2FR method with the GWO optimization algorithm has achieved satisfactory results compared to other classifier methods. Finally, the proposed classification technique was able to provide 72.71% accuracy.

Motivation & Objective

• To address the challenge of subjective, time-consuming, and error-prone clinical diagnosis of SZ and ADHD by developing an automated, objective, and accurate computer-aided diagnosis system (CADS).
• To improve feature representation from rs-fMRI data using a novel 2D convolutional autoencoder (CNN-AE) architecture with 7 convolutional layers.
• To introduce a new interval type-2 fuzzy regression (IT2FR) classifier optimized via metaheuristic algorithms for enhanced classification robustness in complex, uncertain neuroimaging data.
• To evaluate the proposed method’s performance against multiple state-of-the-art machine learning and fuzzy inference systems on a real rs-fMRI dataset.
• To provide a clinically applicable tool that supports early and accurate diagnosis of SZ and ADHD using non-invasive fMRI imaging.

Proposed method

• The method uses the UCLA rs-fMRI dataset containing data from healthy controls (HC), SZ, and ADHD patients.
• Preprocessing is performed using the FMRIB Software Library (FSL) toolbox to standardize and clean the rs-fMRI data.
• Functional connectivity matrices of size 118×118 are extracted from the preprocessed data and converted into 2D images for input into the CNN-AE.
• A custom 2D CNN-AE with 7 convolutional layers is trained end-to-end to learn hierarchical, compact representations of the functional connectivity patterns.
• The extracted features are classified using a novel interval type-2 fuzzy regression (IT2FR) model, optimized via gray wolf optimization (GWO), genetic algorithm (GA), and particle swarm optimization (PSO).
• Ten-fold cross-validation is applied to ensure robustness and generalization of the classification model.

Experimental results

Research questions

• RQ1 Can a custom 2D convolutional autoencoder effectively extract discriminative features from rs-fMRI functional connectivity matrices for SZ and ADHD classification?
• RQ2 Does the proposed interval type-2 fuzzy regression (IT2FR) model outperform traditional machine learning and type-1 fuzzy classifiers in diagnosing SZ and ADHD from rs-fMRI data?
• RQ3 Which metaheuristic optimization technique—GWO, GA, or PSO—yields the best performance when integrated with IT2FR for this classification task?
• RQ4 How does the proposed CADS compare in accuracy to existing state-of-the-art methods that classify only SZ vs. HC or ADHD vs. HC, particularly when distinguishing all three classes (HC, SZ, ADHD)?
• RQ5 To what extent do the functional connectivity patterns extracted from rs-fMRI data reflect known neurobiological differences in SZ and ADHD?

Key findings

• The proposed IT2FR-GWO classifier achieved the highest classification accuracy of 72.71% among all evaluated methods, including MLP, KNN, SVM, random forest, decision tree, and ANFIS-based models.
• The combination of the 2D CNN-AE for feature extraction and IT2FR-GWO for classification significantly improved diagnostic performance compared to baseline models.
• The proposed method outperformed existing state-of-the-art approaches that classify only two classes (e.g., HC vs. SZ or HC vs. ADHD), particularly since it distinguishes three classes: HC, SZ, and ADHD.
• The use of interval type-2 fuzzy logic effectively captured uncertainty in rs-fMRI data, enhancing robustness in the presence of inter-subject variability and noisy signals.
• The functional connectivity patterns extracted via the CNN-AE model revealed significant differences in brain network connectivity between HC, SZ, and ADHD patients, consistent with prior neuroimaging literature.
• The results demonstrate that integrating deep learning with advanced fuzzy inference and metaheuristic optimization is a promising path for automated, accurate, and reliable diagnosis of complex brain disorders.