Brain functional connectivity analysis of fMRI-based Alzheimer's disease data

The prevalence of Alzheimer's disease (AD) poses a significant public health challenge. Distinguishing AD stages remains a complex process due to ambiguous variability within and across AD stages. Manual classification of such multifaceted and massive data of brain volumes is operationally inefficient and vulnerable to human errors. Here, we propose a precise and systematic framework for AD stages classification. The core of this framework discovers and analyzes functional connectivity among regions of interest (ROIs) of a human brain. Multivariate Pattern Analysis (MVPA) is applied to extract features that reveal complex functional connectivity patterns in the brain. These features are then used as inputs for an Extreme Learning Machine (ELM) model to classify AD stages. The model's performance is assessed through comprehensive evaluation metrics to ensure robustness and reliability. Applying this framework on datasets which contain meticulously validated fMRI scans such as the OASIS and AD Neuroimaging Initiative datasets, we validate the merit of this proposed work. The framework's results show improvement in the collective performance of two-class and multi-class classification. Feeding ELM with MVPA features yield decent outcomes given a generalizable and computationally-efficient model. This study underscores the effectiveness of the proposed approach in accurately distinguishing AD stages, offering potential improvements in AD and AD stages detection.

1 Introduction

Alzheimer's disease (AD) is a progressive and irreversible brain disorder that slowly destroys memory and thinking capabilities, ultimately causing loss of cognitive function. It is one of the diseases that cause brain damage (1). The disease typically begins with mild memory loss and may be mistaken for normal aging. As it progresses, individuals may experience significant cognitive impairment, difficulty with language, problem-solving challenges, and a loss of the ability to carry out daily activities (2)

The buildup of tau tangles and amyloid-beta (A) protein in the brain are two of the main symptoms of AD. This neurodegenerative brain ailment progressively impairs cognition in its initial stages and eventually results in mortality. AD is a progressive condition that worsens over time. There are two subtypes of AD (3):

• Early-onset AD and late-onset AD Uncommon genetic abnormalities characterize early-onset AD in patients under the age of 65 (5%).

• AD with a late onset, which affects people with an age onset older than 65 years old (95%) and is not caused by inherited mutations.

Advanced neuroimaging techniques, including magnetic resonance imaging (MRI) and positron emission tomography (PET), allow researchers to detect structural and molecular markers of AD in a non-invasive way, offering both structural and functional insights without the need for physical intervention. Meanwhile, methods like cerebrospinal fluid (CSF) analysis and blood biomarker assessment, though invasive, provide crucial molecular information that can aid in early diagnosis and tracking of AD progression. As neuroimaging technologies continue to evolve at a rapid pace, the volume and complexity of multi-modal data generated are also increasing, posing challenges for data integration, interpretation, and management (4) (see Figure 1).

Figure 1

Figure 1. Invasive and non-invasive approaches.

AD cannot be cured at this time. Some medications can be helpful in symptom control; however, these treatments cannot stop the course of the disease. Unfortunately, the experiments that were conducted to test various strategies, Drugs that diminish the amount of the A peptide or encourage its clearance via active or passive vaccination, for example, failed to show therapeutic efficacy. There have been several tried-and-true methods, such as medications that reduce the amount of A protein or encourage the elimination of the through active or passive vaccination (5, 6).

To address these challenges, computer-aided machine learning techniques have become essential in processing and analyzing neuroimaging data efficiently and accurately. Such techniques enable the extraction of subtle patterns and predictive markers from large datasets, which may otherwise be undetectable to the human eye. Early identification of AD and suitable treatment intervention are crucial aspects of care, as the illness can start years or even decades before dementia develops. As a result, there is a growing focus being placed on carrying out clinical tests in groups consisting of people with no persons with mild cognitive impairment (MCI) or dementia who are in danger of acquiring AD.

1.1 Research contributions

This recent systematic review highlighted the limited number of studies employing MVPA in Alzheimer's research (7). The scarcity was attributed to the complexity and technical challenges of MVPA, including the need for specialized knowledge and difficulties in data preprocessing, feature selection, and classification. Despite the numerous algorithms in the literature demonstrating high accuracy, our focus on MVPA is justified by its unique capability to compute the connectivity between brain regions, offering insights into the underlying patterns of neural activity that other methods might overlook. Our approach aims to enhance the understanding of AD by leveraging the advanced analytical power of MVPA, which could complement and potentially surpass existing methods in capturing the intricate network interactions in the brain.

The key focus areas of this research include:

• Understanding AD through brain connectivity patterns.

• Mapping fMRI scans to a parcellation model using the AAL3 brain atlas to investigate brain connectivity patterns using Machine learning.

• Defining a framework model using Multivariate Pattern Analysis (MVPA) for classification using ELM classifier.

• Evaluate the model using performance metrics by comparing with OASIS and ADNI datasets.

The remaining papers are arranged as follows: offers a comprehensive review of the literature on AD in Section 2, Section 3 The Theoretical Background of AD and brain imaging are explained, Section 4 details our methodology, including Data Information, Preprocessing Pipeline, Functional Connectivity, and Classification Algorithms for different stages. Section 5 presents the discussion, while Section 6 covers the conclusions and future directions.

2 Previous studies

Because early illness identification is so important, numerous scientists are interested in returning to this field to find a therapy for AD. Thus, the most essential works in this field will be presented in this part.

The Graph Convolutional Network (GCN) model, which was established in this study (8), has two main functions: it can be used to diagnose MCI by differentiating MCI from normal controls (NCs), and it can also be used to predict the risk of dementia by categorizing NCs from participants in the OASIS-3 dataset. Possibly having MCI but not yet receiving a formal diagnosis. The findings show that the recommended GCN outperforms both baseline GCN and Support Vector Machine (SVM), with a peak accuracy of 91.2% and an average accuracy of 80.3%, which is 23.5% higher than SVM and 11.7% higher than baseline GCN. Although individual FC often outperforms global FC, under the GCN design, global graphs with the right connections can occasionally achieve similar or even greater performance than individual graphs. GCN categorization is influenced by brain network connections like the default mode, visual, ventral attention, and somatomotor networks. In this study, Alarjani (9) identified key features in fMRI data using a patch-based approach (PBA) and two 3D-CNNs with fully connected layers. We applied decision tree and k-nearest neighbor classifiers to detect AD, MCI, and NC using the ADNI dataset, achieving improved classification performance compared to previous studies.

Kolla et al. (10) used data-driven analysis of resting-state fMRI data to examine functional brain nodes and their effects on graph metrics. It introduces a new measure known as “node-metric coupling” (NMC), which considers the connection between graph metrics and node sizes. The findings suggest that NMC may serve as a valuable biomarker for brain disorders, underscoring the critical need to account for variations in node sizes in neuroimaging studies.

In order to improve the diagnosis of moderate cognitive impairment (MCI), HE et al. (11) employed a graph convolutional neural network to extract information from functional brain networks automatically. GCN with multimodal connectivity, integrates diverse connectivity patterns, achieving 92.2% accuracy and an AUC of 0.988 on the ADNI dataset, outperforming previous methods.

Ĩsmail et al. (12) proposed a novel model combining Capsule Network (CapsNet) and Recurrent Neural Network (RNN) to capture the full 4D information of fMRI data for AD diagnosis. The trial's results demonstrate its effectiveness. Compared to normal control (NC), AD was identified with 94.5% accuracy, while 61.8% accuracy separated early eMCI from late moderate cognitive impairment (lMCI).

This study used the edge time series paradigm in network neuroscience to examine the connections between brain and behavior in AD and related diseases. An examination of data from 152 persons examined the connections between neuropsychological processes and functional brain networks. New connections between attention, limbic, front parietal, and default mode networks and linguistic, executive, cognitive, and attention processes were found by the study. These results demonstrate how edge time-series analysis can go beyond conventional functional connectivity techniques to reveal brain dynamics associated to illness (13).

Kadhim et al. (14) created a computer-aided brain diagnostic (CABD) system capable of detecting AD from brain scans. By employing feature extraction techniques along with fMRI and Positron-Emission Tomography methods, the AD Neuroimaging Initiative (ADNI) data is analyzed. The developed method demonstrates an impressive classification accuracy rate of 97.7% for ADNI images.

Amini et al. (15) developed a machine learning (ML) method for AD diagnosis. To predict the severity of the disorder, they used fMRI data and MMSE scores using CNN, Decision Trees, SVM, KNN, RF, and LDA. Their CNN approach achieved an efficiency of 96.7% and sensitivity rates of 98.1% for low, 95.2% for mild, 89.0% for moderate, and 87.5% for severe AD. These results indicate superior performance compared to previous techniques in detecting AD severity.

Wang et al. (16) created a classification system for fMRI data that differentiates between AD, early and late moderate cognitive impairment (EMCI and LMCI), and healthy controls (HC) using graph theory and machine learning. A dual graph theoretical approach was merged into a unique multi-feature selection technique. Classification accuracy varied from 83.30% to 96.80%, suggesting that graph and ML combined might be useful for diagnosing AD, particularly when local metrics are used to indicate functional changes in specific brain areas. This study explored deep learning for early Alzheimer's detection, focusing on fMRI data after finding genetic factors had little impact. A 3D CNN model achieved 92.8% accuracy using ADNI data, highlighting the potential of deep learning in medical imaging for AD prediction (17).

Alarjani and Almarri (18) applied multivariate pattern analysis (MVPA) to compute connectivity among nodes and to analyze activity across multiple brain voxels. The proposed approach is applied to two public datasets and then classified using various ML classifiers. The results were assessed using performance metrics, and comparisons were conducted to differentiate between various stages of AD.

Table 1 presents a comparative analysis of various studies on AD detection techniques, highlighting their respective methods and limitations. This assessment serves as a valuable reference for understanding the unique attributes and challenges associated with both the graph-based optimization approaches in the context of analyzing brain signals for AD detection.

Table 1

Table 1. Comparison with existing studies.

3 Background

AD progresses through stages: normal, very mild cognitive impairments, mild cognitive impairments, moderate dementia, and severe dementia (see Figure 2). MCI stages are mentioned among dementia and cognitively healthy people. Although MCI patients have lower cognitive function than healthy controls, their indications are less severe than those of Alzheimer's. There is a distinct deterioration in the association between the period with symptoms, MCI, and dementia phases of AD and typical cognitive aging. When someone has MCI, their thinking abilities show minor but significant changes that are visible to them and their family and friends, yet they are still capable of going about their daily lives. MCI affects about (15–20)% of persons 65 and older (19). A sizeable fraction of people can return to their normal cognitive process, whereas others can stay stable and avoid developing AD. The likelihood of acquiring AD is higher in people with antipsychotics and multi-domain MCI, while patients without dementia MCI are more likely to get other kinds of dementia (20, 21).

Figure 2

Figure 2. Phase of MCI stage.

Brain imaging represents a transformative advancement in cognitive neuroscience. Before the advent of imaging technologies, brain research relied heavily on animal models and post-injury evaluations in humans. Diagnosing brain injuries posed significant challenges due to limited tools for assessing internal damage. Modern imaging techniques now provide precise methods for studying the brain, including mapping brain pathways, assessing brain density, and analyzing brain function (19).

Functional magnetic resonance imaging (fMRI) is among the most commonly used techniques for studying brain function, as it can capture high-resolution, real-time images of neural activity. This capability makes it invaluable for understanding both normal and abnormal brain processes. By detecting fluctuations in blood flow and oxygenation levels associated with neuronal activation, fMRI enables researchers to map specific regions of the brain that are engaged during various cognitive, sensory, and emotional tasks. This method provides insights into normal brain function, as well as potential abnormalities linked to neurological and psychiatric conditions. The depth of information fMRI offers makes it invaluable for studying complex neural networks, identifying biomarkers, and tracking subtle changes in brain function over time (22).

This work's primary focus is on computer-aided diagnostics. Achieving accuracy and reliability is crucial for its proper functioning in a healthcare setting, making it one of the most challenging issues in medical image processing. The goal is to categorize patients' cognitive states as CN, MCI, and AD based on their fMRI data. The categorization for AD diagnosis has been the subject of numerous investigations. In additional, He et al. (23) proposed the Spatiotemporal Graph Transformer Network (STGTN) for diagnosing AD using rs-fMRI data. Unlike traditional methods, STGTN captured spatiotemporal features and used adversarial training to enhance robustness and address limited sample sizes. The model achieved high accuracy (92.58% for AD vs. NC and 85.27% for eMCI vs. lMCI), outperforming existing methods. Additionally, it highlighted crucial brain connections for different classification tasks.

Support Vector Machine (SVM) was utilized by Jo et al. (24) to provide an enhanced AD classification system based on MRI indicators for categorizing CN vs. AD SVM was also utilized in Coupé et al. (25) to create a model that relied on the correlation of changes in transverse brain structure—multimodal indicators of MCI development with numerous kernel learning (KCL). However, use a model structure with k-Nearest Neighbors (KNN) (26). Suggested a best-resolution patch-based attributes extraction to retrieve cross-sectional and transverse information from MRI scans of the brain. Several medical fields have used deep learning, including image retrieval, segmentation techniques, image identification, and computer-aided diagnostics.

4 Methodology

The proposed system framework, depicted in Figure 3, consists of several key stages: data collection, data preprocessing, feature engineering, functional connectivity analysis, classification, results generation, and analysis. Each stage is integral to the system's workflow, facilitating the precise and efficient processing of data. The subsequent sections provide a detailed overview of each stage and its role in the framework.

Figure 3

Figure 3. An overview of the proposed framework, including data collection, preprocessing, application of the AAL atlas, connectivity computation, feature selection, model application, and final classification using the ELM classifier.

4.1 Dataset collection and preparation

This study leverages two extensively used neuroimaging databases, the AD Neuroimaging Initiative (ADNI) and the Open Access Series of Imaging Studies (OASIS), which are critical resources in Alzheimer's research. These datasets provide a comprehensive range of imaging and clinical data, enabling detailed analyses of disease progression, early detection, and biomarker discovery.

The rs-fMRI dataset from ADNI was selected for its ability to capture brain activity during resting states without external stimuli. The dataset includes both male and female participants in different stages of AD (27). Specifically, 55 individuals (30 males and 25 females) were selected for the AD stage, 35 individuals (20 males and 15 females) for MCI, and 95 individuals (50 males and 45 females) for NC. Imaging data were collected using a Philips scanner with an 80° flip angle. To address the class imbalance inherent in the dataset, the Synthetic Minority Oversampling Technique (SMOTE) was employed. SMOTE generates synthetic examples for underrepresented classes, effectively balancing the data set and enhancing the performance of the model (28).

For the OASIS dataset, fMRI data were accessed through the official website using verified credentials, which allowed free access to the data. Each 3D image is structured as a volume composed of multiple 2D images. The dataset also includes male and female participants across Alzheimer's stages, with 101 individuals (72 males and 29 females) for the AD stage, 95 individuals (40 males and 55 females) for MCI, and 102 individuals (60 males and 42 females) for NC. These datasets collectively provide a robust foundation for exploring Alzheimer's biomarkers and functional connectivity analysis (29). Figure 4 shows the distribution of subjects across the AD, MCI, and NC stages for the ADNI and OASIS datasets.

Figure 4

Figure 4. fMRI scans of NC, MCI, and AD subjects.

4.2 Preprocessing of fMRI

Since medical images are complex and, hence, hard to extract the features from an image, we need to process images in the dataset using many techniques. All pre-processing techniques are essential for the model during training to make the images clearer, more useful, and perform well. Also, it's necessary during deployment as a part of the model to make the input shape of the images match the input shape of activations from the same brain regions that are accurately captured (30).

Data preparation was conducted using the CONN toolbox (31). The CONN includes several tasks, including segmenting the functional fMRI data, realigning, normalizing, and slicing time correction. It also eliminates noise via de-nosing procedures (32).

4.2.1 Realignment

The second step in our preprocessing pipeline was realignment, aimed at correcting head motion and ensuring that the position of the brain is consistent across all images. This process aligns the functional images so that activations from the same brain regions are accurately captured. In our study, we followed the rational ranges of motion, which were –0.2 to 0.5 for normal controls and –0.2 to –0.15 for individuals with AD (33). As demonstrated in Figure 5.

Figure 5

Figure 5. Image realignment: (A) Each fMRI scan can differ slightly in angle or position because of minor head movements or inconsistencies in the scanner. As a result, the voxels (3D pixels) in each scan may represent slightly different spatial locations in the brain. This misalignment can create inconsistencies when analyzing data from multiple scans, as the same brain region may not appear consistently in the same voxel at different time points. (B) The realignment process modifies each scan to ensure a consistent angle and position, aligning them so that each voxel corresponds to the same spatial location in the brain across all scans.

4.2.2 Registration

Image registration in fMRI refers to the process of aligning multiple images acquired from different time points or subjects to a common reference space. This alignment is crucial for combining or comparing data across different scans or individuals. In fMRI studies, image registration typically involves aligning functional images acquired during the scanning session to a standard anatomical template or to each other (34), as demonstrated in Figure 6.

Figure 6

Figure 6. Image registration: (A) slice without registration (original image). (B) After registrar fMRI with structural MRI.

4.2.3 Slice time correction

The ascending method is commonly employed in slice-time correction to address the temporal misalignment of slices in fMRI data. The slices are obtained interleaved rather than sequentially during the collection of fMRI data. The processing starts with inferior slices and progresses to superior slices for each volume for example, in the Philips slice order, the slices are acquired in 1, 4, 7, 6, 9, etc. Slice time correction reorders these slices to be sequential, which is crucial for proper analysis and interpretation of the data (35), as shown in Figure 7.

Figure 7

Figure 7. Slice time correction inspired by Poldrack et al. (30).

4.2.4 Spatial normalization

SN is a crucial preprocessing step, particularly because brain sizes can vary significantly among individuals. The goal of this procedure is to reduce the total squared deviations between the functional image and a template image that is commonly used. Normalization matches each brain scan to a common standard space, such the Montreal Neurological Institute (MNI), enabling meaningful comparisons across different people and studies. In our study, we employed indirect normalization to MNI space, ensuring that our functional images were successfully transformed into a standardized space for further analysis (36). The normalization procedure is shown in Figure 8.

Figure 8

Figure 8. Spatial normalization.

4.2.5 Spatial smoothing

SS is a crucial preprocessing step in neuroimaging data analysis that raises the signal-to-noise ratio and signal sensitivity. We used a 6 × 6 × 6 mm Gaussian kernel to process the data in our investigation. This kernel size represents the spatial extent over which neighboring voxel values are averaged. By smoothing the data in this manner, we effectively reduce noise and enhance the detect-ability of true underlying signals, improving the quality of subsequent analyses (37). The Smoothing procedure is shown in Figure 9.

Figure 9

Figure 9. Spatial smoothing.

4.3 Feature engineering

Feature engineering is vital in neuroimaging analysis, particularly for AD and other neurological disorders. This process entails selecting and transforming raw neuroimaging data into relevant features that enhance both model performance and interpretability (38).

4.3.1 Brain regions of interest

ROIs are specific brain areas that are selected for analysis based on their known or suspected involvement in a particular process or function. In the context of major brain networks, ROIs are typically selected to represent critical nodes or hubs within these networks. In our paper, we utilized automated anatomical atlas 3 (AAL3) for feature extraction to identify relevant brain regions or patterns in the fMRI data. Following this, extract the time series within each ROI (39). In Figure 10 plotting regions of interest from the AAL3 atlas that includes 170 nodes (i.e., regions). Table 2 represents all brain regions in the AAL3 atlas.

Figure 10

Figure 10. Regions of interest from the AAL3 atlas.

Table 2

Table 2. Name of regions for AAL3 atlas.

4.4 Functional connectivity

Multi-voxel pattern analysis (MVPA) is a neuroimaging technique that examines activity patterns across multiple brain voxels (volumetric pixels). Unlike traditional methods that focus on average activity within specific brain regions, MVPA analyzes the distribution of activity across voxels to decode or classify cognitive states, stimuli, or tasks. This approach offers a significant advantage over single-voxel analyses, as it provides greater sensitivity for identifying specific cognitive states in the brain. By considering the relationships and interactions between voxels, MVPA can detect subtle and distributed patterns of neural activation that might be missed by region-based analysis. This makes MVPA particularly effective in revealing complex brain processes, offering insights into how different brain areas work together to support cognitive functions. Additionally, MVPA has been widely used in studies involving visual perception, language processing, decision-making, and various other cognitive tasks, contributing to a deeper understanding of brain organization and function (40).

Figure 11 presents the time series data for a specific voxel within an fMRI dataset, highlighting the changes in intensity over time. The plot shows the intensity values of the selected voxel at various time points. The x-axis is labeled “TR” (time repetitions), while the y-axis is labeled “Voxel Intensity.”

Figure 11

Figure 11. Plot of voxel intensity over time in fMRI analysis.

In Figures 12, 13 display the dimensions of the correlation matrix, which is the number of ROIs squared. This shows how each pair of ROIs for two distinct datasets is connected to the other.

Figure 12

Figure 12. Functional networks of OASIS dataset: (A) functional connectivity of NC, No. of Subject 002_S_0295. (B) Connectivity of AD, No. of Subject 002_S_0413. (C) Functional connectivity of MCI, No. of Subject 002_S_0729.

Figure 13

Figure 13. Functional networks of ADNI dataset: (A) functional connectivity of NC, No. of Subject OAS30001. (B) Connectivity of AD, No. of Subject OAS30090. (C) Functional connectivity of MCI, No. of Subject OAS30009.

4.5 Classifier

In 2004, Huang et al. (41) introduced the Extreme Learning Machine (ELM), a straightforward yet highly effective single hidden-layer feedforward neural network approach. ELM randomly selects input weights and hidden layer biases, calculates output weights analytically, and overcomes the limitations of conventional single-hidden-layer feedforward neural network (SLFN) learning methods. This technique has found extensive applications across diverse disciplines, including illness diagnosis, traffic sign detection, picture quality evaluation, and beyond (42), as shown in Figure 14.

Figure 14

Figure 14. Architecture of extreme learning machine inspired by Ding et al. (43).

Our proposed model leverages the inherent speed and efficiency of ELM by integrating additional techniques that improve performance on high-dimensional neuroimaging data, especially functional MRI (fMRI). We tackle the complexity of this data using innovative ensemble methods combined with ELM. In this approach, multiple ELM classifiers are trained on various subsets of features or data splits, and their predictions are aggregated. This ensemble method enhances robustness and minimizes the impact of data variability, ultimately boosting classification accuracy. For our Model, we set the parameters as follows: hidden units = 32, activation function = “softmax,” random type = “normal,” and C = 0.1.

4.6 Result and performance validation

The proposed AD classification model is evaluated using multiple metrics, each offering distinct insights into performance. Accuracy measures the ratio of correctly predicted positive and negative cases to the total samples, providing an overall assessment of the model's categorization ability as per (44, 45). However, it may be unreliable for imbalanced datasets as mentioned by Alqahtani et al. (46). To address this limitation as suggest by Pruthviraja et al. (47), Salehi et al. (48), and (49) precision, and recall are also used. Precision evaluates the model's ability to correctly identify positive cases among all predicted positives (50), focusing on prediction accuracy (51). Recall measures the model's sensitivity by identifying all positive cases in the dataset (52). The F1-score, which calculates the harmonic mean of precision and recall, balances these metrics, ensuring a comprehensive evaluation of the model's performance (53).

In Table 3 binary classification tasks, the ELM model demonstrated strong performance, achieving an accuracy rate of 87.80% in distinguishing between NC and MCI, 86.16% between AD/NC, 85.9% MCl/NC, and 83.1% AD/MCL/NC. Also, in term of recall we got 87.10% distinguishing between NC and MCI, 86.3% between AD/NC, 85.9% MCl/NC, 83.7% for multi-class classification (AD/MCL/NC). In term of prec we got 87.3% distinguishing between NC and MCI, 86.57% between AD/NC, 85.8% MCl/NC, 89% for multi-class classification (AD/MCL/NC). In term of f1-score we got 87.90% distinguishing between NC and MCI, 86.22% between AD/NC, 85.9% MCl/NC, 83.20% for AD/MCL/NC for OASIS dataset. In ADNI dataset the binary classification tasks demonstrate the strong performance of the ELM model. It achieved an accuracy of 94% in distinguishing between AD and MCI, 93.9% between AD and NC, 90.6% between MCI and NC, and 94.1% for multi-class classification (AD/MCI/NC). In terms of recall, the model achieved 94% for AD vs. MCI, 93.9% for AD vs. NC, 90.68% for MCI vs. NC, and 94% for multi-class classification. Precision results showed 94.82% for AD vs. MCI, 93.8% for AD vs. NC, 92.9% for MCI vs. NC, and 94.04% for the multi-class task. Finally, the F1-score performance was 94.14% for AD vs. MCI, 93.9% for AD vs. NC, 91.3% for MCI vs. NC, and 94.5% for multi-class classification (AD/MCI/NC). SO, the result of ADNI dataset outperform OASIS.

Table 3

Table 3. Comparison of results from different fMRI scan methods with our work.

Comparing the OASIS and ADNI datasets, the ELM model performs significantly better on the ADNI dataset across all metrics. This suggests the ELM model has a higher generalization capacity and may be more compatible with the ADNI data. These findings imply that the ADNI dataset may offer higher-quality or more discriminative features, which enhances the model's ability to distinguish between different cognitive stages more effectively. The performance differences between the two datasets underscore the importance of selecting the right dataset in machine learning studies focused on AD detection. Furthermore, this indicates that ELM models can leverage specific characteristics of detailed datasets for improved classification results.

4.7 Experimentation setup

All experiments are performed on a PC with 3.7 GHz Intel Core i7-8GB RAM, 250 GB SSD. Programming environment includes 64-bit Windows 10 pro as operating system, MATLAB R2019a, Python 3.9.12, and TensorFlow. Neuroimaging analysis and processing were conducted using the CONN and Statistical Parametric Mapping (SPM) toolbox, which provides comprehensive preprocessing pipelines for fMRI data.

5 Discussion

It is essential to accurately classify MCI as an intermediate stage because this can help in early detection and timely intervention, potentially delaying the progression of AD. Our model's ability to improve classification accuracy at this stage indicates its potential for clinical applications in monitoring cognitive decline. The integration of neuroimaging data, especially functional MRI (fMRI), is crucial for understanding the neural correlates of MCI and AD. This approach enables the identification of specific patterns of brain activity linked to these conditions.

Accessing medical data from local sources is essential for advancing healthcare research and improving patient outcomes. We faced significant challenges in collecting data from local hospitals, mainly due to the scarcity of available local fMRI datasets related to AD. Strict patient privacy regulations and personal concerns about data sharing created obstacles to obtaining the necessary information. While these regulations are vital for protecting patient confidentiality, they can also impede research progress by limiting the availability of relevant datasets.

Changes because of AD, About 100 billion neurons with widespread, growing delays make an adult brain in good health. Individual neurons can connect to other neurons thanks to their expansions. Information is transferred at these interconnections, known as synapses, in the form of brief chemical outbursts emitted by one neuron and picked up by a sensory cell. About 100 of these are further harmed by the brain's diminished capacity so that its primary energy, glucose, is metabolized.

According to studies, the brain alterations linked to Alzheimer's may start 20 years or earlier symptoms manifest. In the early stages of brain alterations, the brain compensates for these changes, enabling individuals to continue functioning normally. However, as neuronal damage progresses, the brain's ability to compensate diminishes, leading to mild cognitive decline. Over time, this decline can interfere with daily activities and impact memory and problem-solving abilities. Recognizing these early signs is crucial for timely intervention and management. Eventually, neuronal loss is so severe that people exhibit overt cognitive decline, comprising signs like memory loss or time- and place-related confusion. Even later, basic physical processes like swallowing become compromised. Although some of the early brain changes related to AD can be recognized in study designs, further study is required to improve the accuracy of the instruments before they can be used in clinical settings. Additionally, although several are being evaluated in clinical studies, there are now no medications that can prevent, slow, or stop these alterations (54)

6 Conclusions

The degenerative and irreversible brain condition known as AD progressively impairs thinking and memory abilities, eventually resulting in the loss of mental function. In this paper, we computed functional connectivity (FC) using Multivariate Pattern Analysis (MVPA). Then we classified fMRI data from the longitudinal OASIS-3 and ADNI datasets into binary and multi-class categories using the Extreme Learning Machine (ELM) model. While our approach demonstrates promising results, several limitations need to be considered. The use of deep learning models, such as the Extreme Learning Machine (ELM), although efficient, introduces challenges. Deep learning models are often viewed as “black boxes,” making it difficult to interpret the features they learn and understand the decision-making process. This lack of interpretability is a significant limitation, especially in clinical settings, where understanding the rationale behind predictions is essential. Future research will employ our techniques to analyze electroencephalography (EEG) recordings of individuals in the resting state (i.e., with their eyes closed) who are suffering from AD, front temporal dementia, and NC (55). We are going to use EEGNet, a specific kind of neural network architecture made to handle EEG data. Using unprocessed fMRI data, our suggested approaches can accurately predict Alzheimer's illness. This research can improve our understanding of the disease and how it progresses, which has enormous clinical relevance.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.

Author contributions

BA: Conceptualization, Methodology, Supervision, Validation, Writing – review & editing. MA: Data curation, Formal analysis, Methodology, Resources, Visualization, Writing – original draft.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This work was supported by the Deanship of Scientific Research, Vice Presidency for Graduate Studies and Scientific Research, King Faisal University, Saudi Arabia [Grant No. KFU250423].

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Gen AI was used in the creation of this manuscript.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

1. Patterson C. World Alzheimer report 2018. Alzheimer's Disease International (ADI) in London, United Kingdom (2018).

Google Scholar

2. López OL, DeKosky ST. Clinical symptoms in Alzheimer's disease. Handb Clin Neurol. (2008) 89:207–216. doi: 10.1016/S0072-9752(07)01219-5

PubMed Abstract | Crossref Full Text | Google Scholar

3. Scheltens P, Blennow K, Breteler MM, De Strooper B, Frisoni GB, Salloway S, et al. Alzheimer's disease. Lancet. (2016) 388:505–17. doi: 10.1016/S0140-6736(15)01124-1

PubMed Abstract | Crossref Full Text | Google Scholar

4. Lombardi G, Crescioli G, Cavedo E, Lucenteforte E, Casazza G, Bellatorre AG, et al. Structural magnetic resonance imaging for the early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment. Cochr Datab System Rev. (2020) 2020:CD009628. doi: 10.1002/14651858.CD009628.pub2

PubMed Abstract | Crossref Full Text | Google Scholar

5. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack Jr CR, Kawas CH, et al. The diagnosis of dementia due to Alzheimer's disease: Recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimer's Dement. (2011) 7:263–9. doi: 10.1016/j.jalz.2011.03.005

PubMed Abstract | Crossref Full Text | Google Scholar

6. Wang H, Yan Z, Yang W, Liu R, Fan G, Gu Z, et al. A strategy of monitoring acetylcholinesterase and screening of natural inhibitors from Uncaria for Alzheimer's disease therapy based on near-infrared fluorescence probem. Sensors Actuat B. (2025) 424:136895. doi: 10.1016/j.snb.2024.136895

Crossref Full Text | Google Scholar

7. Alarjani M, Almarri B. fMRI-based Alzheimer's disease detection via functional connectivity analysis: a systematic review. PeerJ Comput Sci. (2024) 10:e2302. doi: 10.7717/peerj-cs.2302

PubMed Abstract | Crossref Full Text | Google Scholar

8. Han S, Sun Z, Zhao K, Duan F, Caiafa CF, Zhang Y, et al. Early prediction of dementia using fMRI data with a graph convolutional network approach. J Neural Eng. (2024) 21:016013. doi: 10.1088/1741-2552/ad1e22

PubMed Abstract | Crossref Full Text | Google Scholar

9. Alarjani M. Alzheimer's Disease Detection based on Brain Signals using Computational Modeling. In: 2024 Seventh International Women in Data Science Conference at Prince Sultan University (WiDS PSU). IEEE (2024). p. 77–83. doi: 10.1109/WiDS-PSU61003.2024.00030

Crossref Full Text | Google Scholar

10. Kolla S, Falakshahi H, Abrol A, Fu Z, Calhoun VD. Intra-atlas node size effects on graph metrics in fMRI data: implications for Alzheimer's disease and cognitive impairment. Sensors. (2024) 24:814. doi: 10.3390/s24030814

PubMed Abstract | Crossref Full Text | Google Scholar

11. He J, Wang P, He J, Sun C, Xu X, Zhang L, et al. Utilizing graph convolutional networks for identification of mild cognitive impairment from single modal fMRI data: a multiconnection pattern combination approach. Cerebr Cortex. (2024) 34:bhae065. doi: 10.1093/cercor/bhae065

PubMed Abstract | Crossref Full Text | Google Scholar

12. Ĩsmail A, Dalveren GGM. Use of 3D-CAPSNET and RNN models for 4D fMRI-based Alzheimer's disease pre-detection. Turkish J Sci Technol. (2024) 19:223–235. doi: 10.55525/tjst.1396312

Crossref Full Text | Google Scholar

13. Chumin EJ, Cutts SA, Risacher SL, Apostolova LG, Farlow MR, McDonald BC, et al. Edge time series components of functional connectivity and cognitive function in Alzheimer's disease. Brain Imaging Behav. (2024) 18:243–55. doi: 10.1007/s11682-023-00822-1

PubMed Abstract | Crossref Full Text | Google Scholar

14. Kadhim KA, Mohamed F, Najjar FH, Salman GA. Early diagnose Alzheimer's disease by convolution neural network-based histogram features extracting and canny edge. Baghdad Sci J. (2024) 21:0643–0643. doi: 10.21123/bsj.2024.9740

Crossref Full Text | Google Scholar

15. Amini M, Pedram M, Moradi A, Ouchani M. Diagnosis of Alzheimer's disease severity with fMRI images using robust multitask feature extraction method and convolutional neural network (CNN). Comput Math Methods Med. (2021) 2021:1–15. doi: 10.1155/2021/5514839

PubMed Abstract | Crossref Full Text | Google Scholar

16. Wang L, Sheng J, Zhang Q, Zhou R, Li Z, Xin Y, et al. Functional brain network measures for Alzheimer's disease classification. IEEE Access. (2023) 11:111832–45. doi: 10.1109/ACCESS.2023.3323250

Crossref Full Text | Google Scholar

17. Zhou X, Kedia S, Meng R, Gerstein M. Deep learning analysis of fMRI data for predicting Alzheimer's disease: a focus on convolutional neural networks and model interpretability. PLoS ONE. (2024) 19:e0312848. doi: 10.1371/journal.pone.0312848

PubMed Abstract | Crossref Full Text | Google Scholar

18. Alarjani M, Almarri B. Multivariate pattern analysis of medical imaging-based Alzheimer's disease. Front Med. (2024) 11:1412592. doi: 10.3389/fmed.2024.1412592

PubMed Abstract | Crossref Full Text | Google Scholar

19. Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med. (2004) 256:183–94. doi: 10.1111/j.1365-2796.2004.01388.x

PubMed Abstract | Crossref Full Text | Google Scholar

20. Busse A, Hensel A, Guhne U, Angermeyer M, Riedel-Heller S. Mild cognitive impairment: long-term course of four clinical subtypes. Neurology. (2006) 67:2176–85. doi: 10.1212/01.wnl.0000249117.23318.e1

PubMed Abstract | Crossref Full Text | Google Scholar

21. Korolev IO. Alzheimer's disease: a clinical and basic science review. Med Stud Res J. (2014) 4:24–33.

Google Scholar

22. Matthews PM, Jezzard P. Functional magnetic resonance imaging. J Neurol Neurosurg Psychiat. (2004) 75:6–12.

Google Scholar

23. He P, Shi Z, Cui Y, Wang R, Wu D, Initiative ADN, et al. A spatiotemporal graph transformer approach for Alzheimer's disease diagnosis with rs-fMRI. Comput Biol Med. (2024) 178:108762. doi: 10.1016/j.compbiomed.2024.108762

PubMed Abstract | Crossref Full Text | Google Scholar

24. Jo T, Nho K, Saykin AJ. Deep learning in Alzheimer's disease: diagnostic classification and prognostic prediction using neuroimaging data. Front Aging Neurosci. (2019) 11:220. doi: 10.3389/fnagi.2019.00220

PubMed Abstract | Crossref Full Text | Google Scholar

25. Coupé P, Eskildsen SF, Manjón JV, Fonov VS, Pruessner JC, Allard M, et al. Scoring by nonlocal image patch estimator for early detection of Alzheimer's disease. NeuroImage. (2012) 1:141–152. doi: 10.1016/j.nicl.2012.10.002

PubMed Abstract | Crossref Full Text | Google Scholar

26. Kang L, Jiang J, Huang J, Zhang T. Identifying early mild cognitive impairment by multi-modality MRI-based deep learning. Front Aging Neurosci. (2020) 12:206. doi: 10.3389/fnagi.2020.00206

PubMed Abstract | Crossref Full Text | Google Scholar

27. Alzheimer's Disease Neuroimaging Initiative. Available at: https://ida.loni.usc.edu/login.htm (accessed November 01, 2022).

Google Scholar

28. Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE synthetic minority over-sampling technique. J Artif Intell Res. (2002) 16:321–57. doi: 10.1613/jair.953

Crossref Full Text | Google Scholar

29. OASIS Brains Open Access Series of Imaging Studies. Available at: https://www.oasis-brains.org/.htm (accessed December 01, 2022).

Google Scholar

30. Poldrack RA, Mumford JA, Nichols TE. Handbook of Functional MRI Data Analysis. Cambridge: Cambridge University Press. (2011). doi: 10.1017/CBO9780511895029

Crossref Full Text | Google Scholar

31. Nieto-Castanon A, Whitfield-Gabrieli S. CONN Functional Connectivity Toolbox (RRID: SCR_009550), Version 21. Boston: MIT (2021). doi: 10.56441/hilbertpress.2161.7292

Crossref Full Text | Google Scholar

32. Whitfield-Gabrieli S, Nieto-Castanon A. Conn: a functional connectivity toolbox for correlated and anticorrelated brain networks. Brain Connect. (2012) 2:125–41. doi: 10.1089/brain.2012.0073

PubMed Abstract | Crossref Full Text | Google Scholar

33. Zafar R, Malik AS, Kamel N, Dass SC. Importance of realignment parameters in fMRI data analysis. In: 2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA). IEEE (2015). p. 546–550. doi: 10.1109/ICSIPA.2015.7412251

Crossref Full Text | Google Scholar

34. Jonsson T, Wennerberg AB, Forssberg H, Glover GH Li TQ. An image registration strategy for multi-echo fMRI. J Magnet Reson Imag. (1999) 10:154–8. doi: 10.1002/(SICI)1522-2586(199908)10:2<154::AID-JMRI7>3.3.CO;2-#

Crossref Full Text | Google Scholar

35. Parker DB, Razlighi QR. The benefit of slice timing correction in common fMRI preprocessing pipelines. Front Neurosci. (2019) 13:465275. doi: 10.3389/fnins.2019.00821

PubMed Abstract | Crossref Full Text | Google Scholar

36. Calhoun VD, Wager TD, Krishnan A, Rosch KS, Seymour KE, Nebel MB, et al. The impact of T1 versus EPI spatial normalization templates for fMRI data analyses. Hum Brain Mapp. (2017) 38:5331–5342. doi: 10.1002/hbm.23737

PubMed Abstract | Crossref Full Text | Google Scholar

37. Mikl M, Mareček R, Hluštík P, Pavlicová M, Drastich A, Chlebus P, et al. Effects of spatial smoothing on fMRI group inferences. Magnet Reson Imag. (2008) 26:490–503. doi: 10.1016/j.mri.2007.08.006

PubMed Abstract | Crossref Full Text | Google Scholar

38. Sharaev M, Andreev A, Artemov A, Burnaev E, Kondratyeva E, Sushchinskaya S, et al. Pattern recognition pipeline for neuroimaging data. In: Artificial Neural Networks in Pattern Recognition: 8th IAPR TC3 Workshop, ANNPR 2018, Siena, Italy, September 19-21, 2018, Proceedings 8. Springer (2018). p. 306–319. doi: 10.1007/978-3-319-99978-4_24

Crossref Full Text | Google Scholar

39. Rolls ET, Huang CC, Lin CP, Feng J, Joliot M. Automated anatomical labelling atlas 3. Neuroimage. (2020) 206:116189. doi: 10.1016/j.neuroimage.2019.116189

PubMed Abstract | Crossref Full Text | Google Scholar

40. Knyazev G, Savostyanov A, Rudych P, Bocharov A. Multi-voxel pattern analysis of fMRI data during self-and other-referential processing. IP Pavlov J Higher Nervous Activity. (2023) 73:242–55. doi: 10.31857/S0044467723020065

Crossref Full Text | Google Scholar

41. Huang GB, Zhu QY, Siew CK. Extreme learning machine: a new learning scheme of feedforward neural networks. In: 2004 IEEE International Joint Conference on Neural Networks (IEEE Cat. No. 04CH37541). IEEE (2004). p. 985–990.

Google Scholar

42. Huang G, Huang GB, Song S, You K. Trends in extreme learning machines: a review. Neural Netw. (2015) 61:32–48. doi: 10.1016/j.neunet.2014.10.001

PubMed Abstract | Crossref Full Text | Google Scholar

43. Ding S, Xu X, Nie R. Extreme learning machine and its applications. Neural Comput Applic. (2014) 25:549–56. doi: 10.1007/s00521-013-1522-8

Crossref Full Text | Google Scholar

44. Fang K, Chen J, Zhu H, Gadekallu TR, Wu X, Wang W. Explainable-AI-based two-stage solution for WSN object localization using zero-touch mobile transceivers. Sci China Inf Sci. (2024) 67:170302. doi: 10.1007/s11432-023-3968-9

Crossref Full Text | Google Scholar

45. Basheer S, Bhatia S, Sakri SB. Computational modeling of dementia prediction using deep neural network: analysis on OASIS dataset. IEEE Access. (2021) 9:42449–62. doi: 10.1109/ACCESS.2021.3066213

Crossref Full Text | Google Scholar

46. Alqahtani N, Alam S, Aqeel I, Shuaib M, Mohsen Khormi I, Khan SB, et al. Deep belief networks (DBN) with IoT-based Alzheimer's disease detection and classification. Appl Sci. (2023) 13:7833. doi: 10.3390/app13137833

Crossref Full Text | Google Scholar

47. Pruthviraja D, Nagaraju SC, Mudligiriyappa N, Raisinghani MS, Khan SB, Alkhaldi NA, et al. Detection of Alzheimer's disease based on cloud-based deep learning paradigm. Diagnostics. (2023) 13:2687. doi: 10.3390/diagnostics13162687

PubMed Abstract | Crossref Full Text | Google Scholar

48. Salehi W, Baglat P, Gupta G, Khan SB, Almusharraf A, Alqahtani A, et al. An approach to binary classification of Alzheimer's disease using LSTM. Bioengineering. (2023) 10:950. doi: 10.3390/bioengineering10080950

PubMed Abstract | Crossref Full Text | Google Scholar

49. Saleh AW, Gupta G, Khan SB, Alkhaldi NA, Verma A. An Alzheimer's disease classification model using transfer learning Densenet with embedded healthcare decision support system. Dec Anal J. (2023) 9:100348. doi: 10.1016/j.dajour.2023.100348

Crossref Full Text | Google Scholar

50. Si X, He H, Yu J, Ming D. Cross-subject emotion recognition brain computer interface based on fNIRS and DBJNet. Cyborg Bionic Syst. (2023) 4:45. doi: 10.34133/cbsystems.0045

PubMed Abstract | Crossref Full Text | Google Scholar

51. Jia T, Sun J, McGeady C, Ji L, Li C. Enhancing brain computer interface performance by incorporating brain-to-brain coupling. Cyborg Bionic Syst. (2024) 5:116. doi: 10.34133/cbsystems.0116

PubMed Abstract | Crossref Full Text | Google Scholar

52. Lin Y, Chen C, Ma Z, Sabor N, Wei Y, Zhang T, et al. Emulation of brain metabolic activities based on a dynamically controllable optical phantom. Cyborg Bionic Syst. (2023) 4:47. doi: 10.34133/cbsystems.0047

PubMed Abstract | Crossref Full Text | Google Scholar

53. Luo H, Xiang Y, Qu X, Liu H, Liu C, Li G, et al. Apelin-13 suppresses neuroinflammation against cognitive deficit in a streptozotocin-induced rat model of Alzheimer's disease through activation of BDNF-TrkB signaling pathway. Front Pharmacol. (2019) 10:395. doi: 10.3389/fphar.2019.00395

PubMed Abstract | Crossref Full Text | Google Scholar

54. Mueller SG, Weiner MW, Thal LJ, Petersen RC, Jack C, Jagust W, et al. The Alzheimer's disease neuroimaging initiative. Neuroimaging Clin N Am. (2005) 15:869. doi: 10.1016/j.nic.2005.09.008

PubMed Abstract | Crossref Full Text | Google Scholar

55. Doe J, Smith J, Johnson A. A dataset of 88 EEG recordings from: Alzheimer's disease, Frontotemporal dementia and Healthy subjects. (2024). Available at: https://nemar.org/dataexplorer/detail?dataset_id=ds004504&utm_source

Google Scholar