jpad journal

AND option

OR option

ALZHEIMER’S DISEASE PREDICTION USING FLY-OPTIMIZED DENSELY CONNECTED CONVOLUTION NEURAL NETWORKS BASED ON MRI IMAGES

 

R. Sampath1,2, M. Baskar3

 

1. Research Scholar, Department of Computer Science and Engineering, School of Computing, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chengalpattu, Tamilnadu, India; 2. Assistant Professor, Department of Information Technology, Sri Sairam Institute of Technology Chennai, Tamilnadu, India; 3. Associate Professor, Department of Computing Technologies, School of Computing, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chengalpattu, Tamilnadu, India

Corresponding Author: M. Baskar, Associate Professor, Department of Computing Technologies, School of Computing, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chengalpattu, Tamilnadu, India, baskarmsrm@gmail.com, baaskarcse@gmail.com

J Prev Alz Dis 2024;
Published online March 26, 2024, http://dx.doi.org/10.14283/jpad.2024.66

 


Abstract

Alzheimer’s is a degenerative brain cell disease that affects around 5.8 million people globally. The progressive neurodegenerative disease known as Alzheimer’s Disease (AD), affects the frontal cortex, the part of the brain in charge of memory, language, and cognition. As a result, researchers are utilizing a variety of machine-learning techniques to create an automated method for AD detection. The massive data collected during ROI and biomarker identification takes longer to handle using current methods. This study uses metaheuristic-tuned deep learning to detect the AD-affected region. The research utilizes advanced deep learning and image processing techniques to enhance early and accurate diagnosis of Alzheimer’s disease, potentially enhancing patient outcomes and prompt therapy. The capacity of deep neural networks to extract complex patterns from magnetic resonance imaging (MRI) scans makes them indispensable in the diagnosis of AD since they allow the detection of minor aberrations and complex alterations in brain structure and composition. An adaptive histogram approach processes the collected photographs, and a weighted median filter is used in place of the noisy pixels. The next step is to identify the issue region using a deep convolution network-based clustering segmentation process. A correlated information theory approach is used to extract various textural and statistical features from the separated regions. Lastly, the selected features are probed by the fly-optimized densely linked convolution neural networks. The method surpasses state-of-the-art techniques in sensitivity (15.52%), specificity (15.62%), accuracy (9.01%), error rate (11.29%), and F-measure (10.52%) for recognizing AD-impacted regions in MRI scans using the Kaggle dataset.

Key words: Alzheimer’s disease, adaptive histogram approach, deep convolution network-based clustering segmentation, fly optimized densely connected convolution neural networks.


 

Introduction

A brain disorder known as Alzheimer’s disease (AD) causes progressive memory loss in an affected person. Moreover, AD victim’s cerebral capacities are destroyed, which results in catastrophic injury to brain cells. This AD disease occurs in persons with abnormal protein built-up in their bodies (1). Diseases like Alzheimer’s impact millions of people all over the globe and pose a serious threat to modern healthcare. Deterioration of cognitive function, memory, and everyday activities is a hallmark of Alzheimer’s disease, a neurodegenerative disease that worsens with time. The accurate and early detection of AD is still a challenging endeavour, even though medical imaging and machine learning approaches have made substantial progress. Scans using MRI machines are the backbone of modern AD diagnostics because of the priceless information they reveal about the disease’s structural and functional problems. More robust and creative solutions are needed because current systems frequently run into problems with sensitivity, specificity, and efficiency.
A Magnetic Resonance Imaging (MRI) scan detects AD in a patient. The MRI can detect the fundamental causes and characteristics of the illnesses, so it is superior to other diagnostic techniques (2). During the process of detection of AD, the MRI scans will extract the necessary features. The Recurrent Neural Network (RNN) is the most common method to detect AD illness. This RNN makes a clear understanding of the matching that occurs between MRI images and the health condition of the victims (3). A Support Vector Machine (SVM) algorithm is also used to detect this AD in patients. The SVM analyses the MRI scans for any information to identify the reason for the AD via the combined procedures of optimization and classification. SVM is more beneficial to the patients, as it improves the accuracy of disease detection, boosting service delivery efficiency (4, 5).
Region Of Interest (ROI) is a valuable procedure that reviews the areas required for the diagnosis. ROI provides significant information for various research and diagnostic procedures (6). The ROI identification process identifies the dataset that contains necessary information about an image. The K-Nearest Neighbour (KNN) technique is often used during the detection process of ROIs. For the process of filtering ROI from MRI scan images, the KNN approach uses a filtering technique (7). This technique will effectively eliminate unnecessary features and areas from an image. The fuzzy C- Means (FCM) technique is another approach to identifying ROI (8). This segmentation process will reduce the time to identify and analyze an image. A classification technique is utilized to identify MRI image pixels and patterns. ROI identification frequently uses the automatic segmentation method. The system’s efficiency is improved through diagnosis thanks to the information provided by the ROI identification procedure (9, 10).
The leading applications of ML methods are prediction and detection procedures. The ML methods will effectively boost the system’s efficiency by improving the detection process. The detection process of AD disease utilizes ML methods to identify the current condition of patients (11). The convolutional neural network (CNN) based AD detection process is mostly used in healthcare centers. CNN finds out the actual cause and condition of AD by using ARI images (12). Classification and identification processes are used in CNN to identify the important information about AD. CNN trains the dataset that is identified from MRI images (13). CNN employs a feature extraction method to identify the relevant aspects of an MRI scan. Details extracted during feature extraction are then segmented using the segmentation method in CNN. The efficiency of medical facilities is increased by the CNN-based AD detection approach’s high detection rate. CNN decreases the detection process’s error rate and latency (14, 15).
The capacity of deep neural networks, and specifically CNNs, to analyze intricate patterns in medical imaging data makes them indispensable in the assessment of AD. CNNs are very good at recognizing patterns, so they can pick up on MRI scans that show signs of Alzheimer’s disease, even if they’re little. They can streamline the diagnostic procedure by automatically extracting relevant features from raw data, decreasing reliance on manual interpretation. Increased accuracy and practicality in AD detection are outcomes of its scalability, which permits training on massive datasets. The advantages of automated methods include high precision, reliability, efficiency, rapidity, scalability, and availability. The study’s unique use of deep convolutional neural network-based clustering segmentation, employment of advanced image processing techniques, and novel integration of metaheuristic optimization with deep learning all contribute to the advancement of AD diagnosis utilizing MRI data. As a result of this study’s findings, our understanding of Alzheimer’s disease (AD) and how to treat and care for patients with this devastating neurological disorder will significantly improve.
The purpose of this research is to use MRI data to create an automated method for the early diagnosis of Alzheimer’s disease. The technique will examine MRI scans of the brain for signs of Alzheimer’s disease. Modern deep learning and machine learning methods will be used to fine-tune the system. Massive MRI image datasets including both patients and healthy controls will be used to verify the method. The method will be evaluated in comparison to current methods for Alzheimer’s disease diagnosis, showing how it is better. The goal of the study is to evaluate the method’s effect on patient outcomes, diagnostic precision, and treatment effectiveness once it is put into clinical practice. This study improves the detection of Alzheimer’s disease using MRI pictures by combining metaheuristic optimization with deep learning approaches. The adaptive histogram method is employed to enhance picture quality by dealing with noisy pixels. Additionally, the approach makes use of clustering segmentation based on deep convolution networks and feature extraction methods. This strategy is vital for clinical settings where correct diagnosis is necessary for patient management and treatment planning. Experimental validation using MRI scans indicated that it reduces error rates and enhances detection accuracy.
The main contributions of this study is:
– To develop a comprehensive framework for automated Alzheimer’s disease diagnosis using MRI images, integrating advanced techniques like fly-inspired optimization algorithms, deep neural networks, adaptive histogram methods, and weighted median filters.
– To implement an innovative image processing techniques like adaptive histogram methods and weighted median filters to improve the quality and clarity of MRI images, thereby enhancing the accuracy of diagnosis and reducing false positives.
– To employ the fly-inspired optimization algorithms to optimize the architecture and parameters of convolutional neural networks for Alzheimer’s disease diagnosis, enhancing the performance of automated diagnostic systems and providing valuable tools for early detection and intervention.

The rest of the paper is organized as follows: section 2 discusses about various related studies in the literature. Section 3 describes about the proposed methodology in detail. The results and discussion on this study is presented in section 4. Finally the paper concludes in conclusion section.

 

Related Works

Patients with Alzheimer’s disease, corticobasal degeneration, globular glial tauopathy, Pick disease, and progressive supranuclear palsy were used by Kim et al., (16) to train a poorly supervised deep learning model for tauopathy diagnosis. The model obtained the best AUC and diagnostic accuracy of 0.873 ± 0.087 by utilizing clustering-constrained-attention multiple-instance learning (CLAM). However, there is a lack of research on clinicopathologic correlations, new characteristics and imaging modalities, the generalizability of CLAM, and how well models function in large-scale studies.
Guo et al. (17) proposed an Enhanced Deep Learning Algorithm (EDLA) to identify the AD disease. The disease’s most prominent characteristics and patterns are identified by employing the state functional data. EDLA utilizes the method of a neural network to mark the real source of AD and provide a valuable dataset for clinical assessment. The proposed method minimizes the time taken to diagnose AD.
The DANMLP methodology, established by Qiang et al. (18) and employing multi-modal data such as sMRI, clinical data, and APOE genetic data, has demonstrated a notable enhancement in the accuracy of Alzheimer’s disease and mild cognitive impairment classifications. It gets 82.4% accuracy in the ADNI database and 93% accuracy in the 5CV test. While the DANMLP’s personalized visualization helps doctors detect Alzheimer’s disease early on, the tool still needs work in areas including interpretability, data mode integration, and validation of performance across different demographics and datasets.
Alinsaif et al. (19) proposed a 3D approach to shearlet-based AD disease detection. Data for detection are obtained from the images captured using MRI scans, which are given as input. To investigate the sickness, the suggested methods employ a CNN technique. It provides a classification model that divides the data and their characteristics into subcategories based on a predetermined set of rules. The proposed approach increases the trustworthiness and practicability of medicinal analysis in healthcare facilities.
Bai et al. (20) introduced a new approach called Brain Slice Generative Adversarial Network (BSGAN) to detect AD (ADD) disease. The suggested approach uses a generator to identify the disease feedback and classifications that give the most useful data for ADD. The unique characteristics of the brain can be identified by the BSGAN approach, which can illuminate the disease. The efficiency is improved to its maximum potential by the proposed BSGAN, thereby increasing the ADD process’s accuracy.
Mehmood et al. (21) described the transfer learning strategy that aided the identification of AD disease. The proposed methodology is employed at the initial stages of AD diagnosis. This method utilizes tissue segmentation to determine the location and health of brain cells. The dependability and efficiency of the system are increased by the proposed approach that can utilize its characteristics of decreased latency rate during the detection process.
Zhang et al. (22) introduced a unique Generative Adversarial Network (GAN) model that was constructed over Positron Emission Tomography (PET) to assist AD detection. PET captures all the relevant information regarding the brain cells in this approach. The proposed approach increases the speed of calculations and thereby reduces the effort and time required for the calculation that can be performed. It is recommended that the BPET-GAN model performs better than other existing models for detection.
Ebrahimi et al. (23) proposed a new AD detection approach employing an MRI model. Features are extracted, in this case, using the CNN technique. The information generated by MRI scans regarding categorization and identification is unparalleled. A feature extraction step produces an authentic information collection for detection by isolating relevant patterns and characteristics. The suggested approach has a high rate of disease detection, thereby increasing the effectiveness of diagnosis.
Feng et al. (24) presented a novel AD detection technique based on cerebrum Region Of Interest (ROI). The proposed approach uses a structural MRI (sMRI) to store the anatomical data about the cerebrum. ROI is utilized for the classification of brain frequencies and characteristics that are frequently used for AD detection. The suggested approach reduces the error rate in the detection process, thereby increasing the efficiency and practicability of AD diagnosis.
Guan et al. (25) proposed a multi-model system to facilitate AD diagnosis based on MRI. The identification and classification procedure delay rate is reduced while AD detection is processed using MRI scan images. The multi-model approach improves the system’s efficiency, thereby reducing the error rate and complexity of the computing process. The analytic results demonstrate that the suggested approach will effectively increase AD diagnosis by achieving a high accuracy rate of detection.
De Souza et al. (26) presented a new computer-aided method for analyzing AD disease. The data collected by the MRI scan images is usable and practical. This paper examined MRI scan images for proof of AD and noted their locations. MRI images are cut using an evolutionary computing system to reduce delay processing in such cases. The proposed approach improves the accuracy and efficiency of AD detection.
Jo et al. (27) proposed an AD detection approach given by Positron Emission Tomography (PET). The suggested approach employs a feature and classification of the extraction procedure for detecting AD using deep learning. In this case, precise characteristics from PET images are extracted using a 3D convolutional neural network (CNN) method. The proposed method increases diagnostic efficiency by optimizing AD detection’s practicability and accuracy.
Ni et al. (28) presented a PET technique for diagnosing Alzheimer’s disease that relies on ethyl cysteine dimers (ECDs). Here, SPET records the neural processes responsible for generating visual detail. Here, ECD is used to SPET pictures to distinguish between cognitive and normal cognitive (NC) data. The proposed ECD-SPET approach increases system efficiency by decreasing detection latency and increasing AD detection accuracy.
Alzheimer’s disease (ADD) detection was given a fresh spin by Chen et al. (29), who advocated using a convolutional neural network (CNN) method. In this case, MRI images are processed through a feature extraction approach to pull out the most relevant sets of characteristics. The proposed solution increases system efficiency by improving the dependability of the calculation process. The proposed ADD method reduces the computation cost and improves the accuracy rate in providing services for the AD diagnosis process.
Helaly et al. (30) presented a deep-learning hippocampal segmentation framework (DL-AHS) for Alzheimer’s disease. The detection process is improved by analyzing data from the left and right hippocampi, as in the proposed method. The segmentation approach precisely noted the patient’s condition to speed up the process’s computation time. The proposed approach increases the accuracy of the detection of diseases and the rate of performance.
Liu et al. (31) presented a neural network model based on Deep Separable Convolutional (DSC) for detecting AD disease. The DSC exploits a feature extraction strategy to zero in the relevant explanation of an MRI scan. The classification process again categorizes the information obtained from the feature extraction process. The suggested approach increases the rate of success in detecting AD. The proposed DSC method reduces the computational effort and time required for various traditional approaches.
Pan et al. (32) proposed a Genetic Algorithm (GA) based on an AD detection model that utilizes three-dimensional CNN (3DCNN). The data obtained from CNN is sorted out by a GA classifier. The classifiers identify the ROI for AD detection and produce the best possible data set. This generated data set by classifiers is then used to practice the 3DCNN model. Using the proposed 3DCNN model, the overall diagnostic performance is improving, and the accuracy of AD detection has increased.
Shahwar et al. (33) introduced a hybrid classical-quantum model of neural networks for detecting AD. The suggested approach provides the required planning information for AD diagnosis based on a Quantum Machine Learning (QML) approach. A feature extraction approach isolates the relevant feature data and vectors to facilitate the detection. The proposed model improves the system’s practicality and accuracy rate by increasing the accuracy rate of the AD detection process.
Tanveer et al. (34) suggested collecting and preparing datasets utilizing ADNI archive MRI scans. Deep Transfer Ensemble (DTE) is a computationally efficient deep neural network ensemble trained to utilize transfer learning for feature extraction. The model is tested on cognitively normal vs AD and moderate cognitive impairment vs NC classifications. Snapshot ensembles and other deep models are less accurate than DTE on large and small datasets. Further research is needed to assess DTE’s robustness, explore ensemble approaches, and examine ethical and regulatory issues.
A methodology was suggested by Özçelik and Altan (35) that utilizes preprocessing 12,500 color fundus images, 96 features were extracted using a two-dimensional stationary wavelet transform, features were selected using k-nearest neighbor and chaotic particle swarm optimization algorithms, and an AI-based classification model was developed. The accuracy and robustness of the model were demonstrated by testing on 2500 photos. However, there is a lack of research on testing performance in real-world clinical contexts, studying how well models generalize across different patient groups and imaging modalities, and considering other feature selection strategies.
Altan and Yağ (36) employed a two-dimensional discrete wavelet transform to extract features from an image dataset containing apple, grape, and tomato plants. The feature selection process makes use of the wrapper approach, which integrates the flower pollination algorithm with a support vector machine. Two optimization algorithms are compared: particle swarm optimization (PSO) and the one that is proposed. The model is tested on plants in real time using an unmanned aerial vehicle. Findings demonstrate good precision with little computing burden; possible areas for further study include investigating metaheuristic optimization methods, testing the model’s scalability, and improving its performance in large-scale farming.
Özçelik and Altan (37) extracted characteristics for a classifier model using 2400 color retinal pictures. The model employs a hybrid strategy for feature selection called wrapper-based k-Nearest Neighbor and genetic algorithm. It outperforms Gaussian Naive Bayes in DR disease stage classification, achieving good accuracy in both early and later stages. The model’s efficacy in clinical contexts, investigating other feature extraction methods, and thinking about ethical and regulatory considerations are all possible areas where research is lacking.
Deep learning algorithms, CNN, and MRI scan segmentation methods are the mainstays of current attempts to diagnose Alzheimer’s disease, although their shortcomings are detailed in the literature review. Reduced accuracy, longer processing times, and complexity are some of the drawbacks of these technologies. The review suggests a new way to diagnose AD that is more efficient, accurate, and scalable, and it tackles these problems head-on. Integrating an adaptive histogram technique with fly-optimized densely linked convolutional neural networks, the suggested method aims to transform AD detection. The study’s overarching goal is to prove that the suggested method is novel and crucial to the development of automated AD diagnosis.

 

Proposed Method

The proposed method detects Alzheimer’s disease-affected patients in healthcare centres through MRI image input. Alzheimer’s is a brain disease called dementia, which typically impacts memory power and other mental abilities to perform daily activities. The existing systems only help with identifying the symptoms of AD; there are no treatments available to cure or stop its progression. This leads to memory loss and outputs in biomarkers, and the ROI of the affected regions does not contain generation in the temporal lobe, parietal lobe, and cingulate gyrus. This study’s proposed custom approach, «Alzheimer’s Disease Prediction Using Fly-Optimized Densely Connected Convolution Neural Networks Based on MRI Images,» is crucial for several reasons. They provide benchmarking and comparison, optimize for particular needs, address constraints, increase scientific understanding, and supply individualized solutions. Researchers can overcome the constraints of current tools and create solutions that are more durable, adaptable, and reliable by using custom approaches. This strategy encourages constant development in the sector by expanding the limits of what is presently achievable. Fig. 1 illustrates the functions of the proposed method.

Figure 1. Proposed Method Functions

 

Metaheuristic algorithms are optimization approaches that are utilized in deep learning to optimize the weights and biases of deep neural networks. These techniques are useful for discovering the best neural network architecture configurations by exploring the solution space. The adaptive histogram approach and the weighted median filter can be used to improve the brightness and contrast of MRI pictures. Clustering segmentation using deep neural networks divides images according to similarity criteria, separating Alzheimer’s disease-affected regions for additional examination. Improve the accuracy, specificity, and sensitivity of densely linked convolutional neural networks for Alzheimer’s disease detection using techniques inspired by the optimization of flying insects. Essential to Alzheimer’s disease diagnosis is feature extraction, which records the disease’s characteristic patterns, textures, and structures. The goal of these methods is to make MRI-based diagnoses of Alzheimer’s disease more precise and trustworthy.
The assimilation of certain features based on textural and statistical is to classify various kinds of MR image input. In this way, the first process is the feature extraction of MR images of the patient’s brain as abnormal or normal. Using large data volume as image features to handle the AD-affected region using metaheuristic optimized deep learning for adaptive histogram approach is achieved. However, the discussed work requires sensitivity, specificity, and accuracy with histogram representation to retain the image features.
This consecutive manner of applying the histogram approach is to improve the image quality, and a weighted median filter replaces the noisy pixels in MRI images. The AD-affected region is identified through a deep convolution network-based clustering segmentation process. Medical images, especially MRI scans used to diagnose Alzheimer’s disease, are improved using the adaptive histogram method and a weighted median filter. The adaptive histogram method enhances picture sharpness and detail by adjusting picture brightness and contrast through redistributing pixel values. The weighted median filter uses pixel-level similarity to assign weights, allowing it to filter out noise while keeping image details intact. The method successfully reduces background noise while maintaining critical anatomical features. Image regions are filtered according to their unique properties, which improves clarity and contrast. Suppressing noise to an acceptable degree without degrading picture quality requires iterative tuning.
The suggested model for image segmentation using DCNNs encounters multiple obstacles. The intricacy of brain architecture, imaging settings, patient movement, and hardware restrictions are a few of these elements that can cause images to have varying degrees of quality. MRI images can also be affected by noise and artifacts. In Alzheimer’s disease, it can be particularly difficult for the model to correctly differentiate between various brain areas. It might be challenging to acquire a big, diversified dataset with annotated ground truth labels for training a DCNN owing to issues with data availability, privacy, and expertise. Another potential issue with DCNN models is overfitting, which makes them underperforming when it comes to segmenting new MRI images that are different from the training set. A large volume of data is required in the Kaggle database. The errors are detected as an instance of identifying the noise pixels. The Kaggle database analysis focuses on Alzheimer’s disease detection through MRI image input. The Kaggle dataset is essential for testing an Alzheimer’s disease diagnosis algorithm. MRI scans grouped by disease stage provide a complete dataset for algorithm training and evaluation. The dataset focuses on Alzheimer’s disease diagnosis, allowing the algorithm to be tested for early indications and stages. Prediction efficacy is measured by sensitivity, specificity, accuracy, error rate, and the F-measure, which combines precision and recall. These parameters show the algorithm’s performance in Alzheimer’s disease detection, including sensitivity to detect true positives, specificity to avoid false positives, accuracy, mistake rates, and precision-recall balance.
Initial MRI image input of MRIi represent the instance of affected brain cells identified at different time intervals such that the feature extraction F(x) is given as


In equation (1), the variable t is used to denote the existing processes consume time and the main goal is to error minimization in MRIi(t) ∈ F(x) is performed. The large volume of data is classified into two instances namely textural features(Tf ) and statistical feature(Sf). Hence,F = Tf+Sf such that Tf is detected between two instances of statistical feature observation or vice-versa. If P represents the number of processing instances, therefore Tf={(P×F)-Sf} is the consecutive process for automatic detection of AD through machine learning techniques. LetHr(Tf,Sf) andHr(Tf) represents the histogram representation of MRIi identified in P at different intervals and t is observed for all Tf such that

Where,


From the above equation (2a) and (2b), the gathered MRI images are performed by applying the histogram approach in {P×(Tf+Sf)} and (t/P Tf) instance to handle large data volume while detecting the biomarkers and ROI with MRIi. Here, based on the histogram representation as in the above equations is rewritten as


From the above-expanded feature extraction process in MRI images, the instance of Tf∈F is previously computed for addressing the first textural feature observation and is derived as in the equation (4). This operation is used for detecting the textural feature observation instance of P∈Tf is given as

In equation (4), the instance of existing processes with large data Tf-1 (i.e.) the existing process sequences of(Tf+Sf) and

Hence, the collected data are processed in the above instance, F(x)=Hr (Tf,Sf)-[1-P(Hr(Tf))] is the last output for t≠0 condition. The initial segmentation and pixel replacement process are illustrated in Figs. 2 and 3 respectively.

Figure 2. Initial Segmentation Process

Figure 3. Pixel Replacement Process

 

The input image is processed for extracting textural and semantic features for the given dimensions. The extracted pixels are identifiable/unidentifiable depending on the density and F(x) value. The histogram representation is modified using equation (3) for unidentified pixels. The unchangeable (due to unidentified) pixels are segmented at the initial stage (Refer to Fig. 2). Improving the quality of MRI images and locating areas impacted by Alzheimer’s disease are two primary goals of the Initial Segmentation Process. The procedure includes preparing the images, extracting features, segmenting them, finding the pixels that cannot be changed, modifying the histogram, and finally, producing segmented output. This procedure aids in the localization of ROIs that may exhibit symptoms indicative of disease, such as wasting away or aberrant protein buildup. To facilitate subsequent processing, immutable pixels are highlighted, and the histogram is adjusted accordingly. All of the analysis and classification that follows in the AD detection pipeline is based on the segmented output. These steps allow for precise localization of AD-affected areas in MRI scans, which is crucial for the AD detection framework. This process replaces the noisy pixels in MRI images based on (NpTf,Sf) and(NpTf) for the segmented region and correlated information instances at the first input image processing is computed as


In this equation, Np (Tf,Sf), represents the estimation of noise pixels in the segmented region of the MRI image. Hr (Tf,Sf), denotes the histogram representation of the segmented region, which captures the distribution of pixel intensities. ∑P∈t[Sf×MRIi]t, calculates the sum of the product of the statistical features Sf and the MRI image MRIi over the time intervals t. Where applying the adaptive histogram approach in MRI image input focuses on identifying the AD-affected region Ar and errors e based on large volume data are estimated as

Here, the equation uses a mix of noise pixels and histogram representations to identify the AD-affected region and estimate the noise pixels in the segmented region. By analyzing histogram representations and noise pixel estimations, the AD-affected region is represented by Ar. The letter e stands for estimating mistakes or inadequacies. Equation 6 calculates the total number of noise pixels over time intervals, same as Equation 5, except it accounts for areas affected by AD and estimating mistakes.
Based on equation (5) and (6) computes the biomarkers and ROI region of AD-affected places based on a textural and statistical feature of the instance at different time intervalst. That is stored from the existing processes of ROI region detection. In this processing, the assessment of NpTf,Sf,NpTf,Hr (Tf,Sf) and Hr(Tf) are the serving input MRI image instances through deep convolution network-based clustering segmentation. Fig. 3 presents the pixel replacement process.
The unchangeable representation (segmented) image is first analyzed for its pixel distribution through classification. In the classification process, independent and noisy pixels are identified. From the independent pixels, the joint pixels are identified for preventing modification(Tf,Sf). If the substitution succeeds, the reconstructed histogram representation is presented (Refer to Fig. 3). Pixel replacement is an essential step in MRI for the diagnosis of diseases, including AD. The process begins with the output of the segmentation and continues with the examination of the distribution of pixels, their classification, and the identification of joint pixels. Independent pixels stand for real tissue characteristics or anatomical structures, whereas noisy pixels might be the result of imaging flaws or errors. Joint pixels may indicate areas impacted by pathological alterations due to their associated or interrelated features. The goal is to enhance the image quality and dependability by replacing noisy pixels. Reducing artifacts and noise in the reconstructed image allows for a more faithful portrayal of the disease and underlying anatomy. An area of cohesive or interconnectedness inside the biological tissue is indicated by joint pixels in independent pixels.

Deep Convolution Neural Network-based Clustering Segmentation

The deep convolution network-based clustering segmentation process is used for identifying the AD-affected region in MRI images based on NpTf,Sf or NpTf and calculating NP ande. The learning process for cluster formation is illustrated in Fig. 4.

Figure 4. Learning Representation for Cluster Formation

 

The inputs are NPTf ∧ Hr (Tf,Sf) for the densely connected convolution network. This is mapped for P(Tf)suchthatHr (Tf)∧Afr are segregated for identifying non-replaceable (noisy) pixels. Depending on the Hr (Tf)*Afr combination, F(x)∈Tf∈P(Tf)∧(Sf+Tf) factors are extracted. This extraction is used for grouping F(x)∈Tf∧F(x)Sj independently as the identified ROI. This is performed to segregate errors and actual clustering inputs (Refer to Fig. 4). The following procedures make use of the fly optimization algorithm:

Step 1: Initialization

Initializes the AD-affected human size(HP) and then the maximum number of portions incorporated based on the brain disease for applying an adaptive histogram approach from the collected MRI image inputMax(HP). Identifies the AD-affected region in the human body based on MRI image for x and y axis based on the fly optimization.

Step 2: Noise pixel-based identification process

Step 2.1: Compute the random biomarkers and the ROI region replaces the noise pixel based on the identification.

Here, X and Y represent the coordinates of the pixels in the image. x-axis and y-axis denote the original coordinates of the pixels in the image. Np represents the noise pixel identification.

Step 2.2: Compute the clustering (C) between each MRI image and the features. Calculate the segmentation value (Sg) of the AD-affected region, which is the reverse of the noisy pixel identification process:

Where, Ci is the clustering between the MRI image and the features. x and y represent the coordinates or distances between the MRI image and the features.
Where,


Here, Sgi is the segmentation value of the AD-affected region and Ci is the clustering value computed using Equation (8).

Step 2.3: The F-measure valueSgi of the affected region is identified based on the histogram representation to calculate the AD-affected Afr a region in the MRI images.

Here, Afr represents the AD-affected region in the MRI images and Sgi is the F-measure value of the affected region. The function detect is used to determine whether the F-measure value Sgi indicates the presence of the AD-affected region.

Step 2.4: Identify the best F-measure value for identifying AD-affected regions in MRI images and corresponding optimal ROI for clustering segmentation:

Where, bestF-measure represents the best F-measure value obtained for identifying AD-affected regions. bestdetection indicates the optimal detection result corresponding to the best F-measure value. Sgi denotes the F-measure value calculated for the affected regions in the MRI images. The function maxdetect is used to determine the maximum F-measure value and its corresponding detection outcome.

Step 3: AD-affected region identification process based on noise pixel identification in MRI images. Information about the ideal F-measure value of the AD-affected area and its coordinate ROI is remembered, and members of the group can then easily locate the ROI:

Such that,

Step 4: Iterative optimization

Iterate through steps 2 and 3 until the maximum number of iterations, maxdetect(Sgi) has passed without a single noise pixel being identified in the ROI and the superior detection value being lost. Fig. 5 presents the flow of the fly optimization algorithm.

Figure 5. Fly Optimization Flow

 

The fly optimization is used to investigate the selective features in MRI images and is computed at the end of all the textural feature extraction or before the initialization process of correlated information based on Tf andSF. Fly optimization approaches are intricate and call for a thorough familiarity with optimization theory and algorithms. Complex mathematical formulas and iterative procedures are involved, necessitating an in-depth familiarity with optimization theory and algorithms. Search strategy, exploration-exploitation harmony, convergence criteria, and initialization procedures are all parameters that must be fine-tuned to achieve optimal performance. When working with big datasets or complicated optimization issues, these techniques tax the computer’s resources. Particularly in non-convex, multimodal fitness landscapes or high-dimensional search spaces, convergence and stagnation might be difficult. Small adjustments to the starting points and parameter settings can cause fly optimization algorithms to take different paths to optimization and provide different results.
The goal of developing fly-optimized densely linked CNNs is to make medical imaging data, especially MRI scans, more reliable, efficient, and accurate for dementia diagnosis. This fly-inspired metaheuristic approach optimizes CNN design and parameters for Alzheimer’s disease diagnosis. CNNs detect regions of interest and modest biomarkers in MRI scans that are suggestive of AD pathology. They accurately identify AD disease by categorizing MRI images into affected and non-affected groups based on learning patterns. Enhancing performance and resilience, the fly optimization technique fine-tunes the CNNs’ architecture and settings. Clinical and research applications of fly-optimized CNNs are supported by their scalability, generalizability, and suitability for use with varied datasets and patient populations.

Experimental Results Tabulation

This subsection details the experimental evaluation outcomes of the suggested approach on the dataset (38, 39). These dataset provides inputs of more than 6400 MRI images diagnosed with AD. The AD is classified as “Mild Demented”, “Moderate Demented”, “Non-Demented”, and “Very Mild Demented”. The training set contains 717, 52, 2560, and 1792 images in the above classification order. The following Tables 1,2 and 3 illustrate the step-by-step process of the proposed method.

Table 1. F(x) based Histogram Improvements

 

Table 2. Pixel Distribution and Replacement

Table 3. Segmented Output

 

Post this analysis, the error from the different solutions achieved is presented in Fig. 6.
Data analysis, pattern recognition, and image processing all benefit from histogram representation for data segmentation and classification. It helps with feature extraction, simplifies complex data, and shows how data is distributed visually. When it comes to classification and segmentation jobs, histograms are your best chance for getting the job done with pinpoint accuracy. By dispersing pixel intensities according to local histogram statistics, adaptive histogram equalization improves contrast. By removing irrelevant pixel values or adjusting the intensity distribution, histogram-based methods can lessen picture noise. Pattern recognition applications also make extensive use of histogram representation.

Figure 6. Error for Different Solutions

 

The results of the performance evaluation of different methods for Alzheimer’s disease detection are displayed in Fig. 6. It allows users to compare error rates by visually representing the errors associated with each solution. Lower mistake rates indicate more accuracy and reliability, as seen in the chart that examines the effectiveness of each solution. This data can be used by researchers to make future versions of detecting algorithms even better. It is used as a standard to compare the effectiveness of new treatments to current approaches for detecting Alzheimer’s disease. Researchers and practitioners can use error analysis to make better decisions when it comes to optimizing and selecting algorithms.
Integrating deep learning, image processing, and statistical analysis, the study seeks to identify AD from MRI inputs. Methods include preparing MRI scans, extracting features, classifying them, evaluating them, and finally optimizing them. The F-measure, sensitivity, and specificity are some of the measures used to evaluate the model’s performance. It might also go through optimization procedures like ensemble learning, hyperparameter tuning, or cross-validation. Errors in training, validation, testing, and optimization are all shown in Fig. 7, which shows the outcomes of the individual analyses. A failure to generalize to new data is indicated by high validation errors, while a high training error rate indicates an inability to learn from training data. While optimization mistakes determine the best way to set up the model, testing errors gauge how well it performs in actual situations.

Figure 7. Individual Analysis

 

The error observed in the training, validation, testing, and best (optimization) is presented in the above figure. The training improves the considerations for pixel classification and hence CNN is densely populated forF(x)∈TfandSf independently. This training requires multiple epochs for stabilizing the validation output to reduce errors. Based on the recurrent process, the mean error is reduced by identifying(Sf+Tf). The optimization process identifies the best fit solution ofF throughmaxdetect(Sgi). This is obtained from the independent analysis presented in Fig. 7. This independent output is modeled using the seriesF=Tf+Sf,Hr (Tf) (before and after modification), andNpTf mitigation.

 

Comparative Analysis

This section presents the comparative analysis results for the metrics’ sensitivity, specificity, accuracy, F-measure, and error rate. The comparisons are performed by varying the features from 2 to 26 and the regions from 1 to 10 as extracted. In the comparative analysis, the methods DSC-NN (31), EC+ML (26), and HCQMLM (33) from the related works section are considered.

Sensitivity

Based on textural and statistical feature extraction utilized to detect the errors, the presented technique provides excellent sensitivity for identifying AD-affected areas in the human body at various periods (Refer to Fig. 8). The large volume of data accessed based on noisy pixel identification severely impacts an individual’s ability to perform daily activities. The collected images are useful for processing automatic AD detection by applying an adaptive histogram approach MRIi(t)∈F(x) based image segmentation process.

Figure 8. Individual Analysis

 

Specificity

The AD-affected region identification in MRI images achieves high specificity for error rate detection based on textural feature instances (Refer to Fig. 9). The memory control, language, and thoughts are managed and controlled based on the conditionTf={(P×F)-Sf} for detecting errors due to existing processes consuming more time. This error is addressed through machine learning and correlated information processed based on the AD-affected region identification through the deep neural network. This consecutive process was validated in each ROI region detection level to reduce the error rate.

Figure 9. Specificity Analysis

 

Accuracy

This fly-optimized densely connected convolution neural network achieves high accuracy compared to the other factors in the user information access (Refer to Fig. 10). In this manuscript, the noisy pixel identification is used for detecting errors in the ROI region through machine learning that affects the brain cells

Figure 10. Accuracy Analysis

 

Based on Alzheimer’s disease detection, the increasing correlated information is identified through deep belief network-based clustering segmentation [as in equation (4)].

F-Measure

In Fig. 11, the noisy pixels and ROI region identification through machine learning techniques detect Alzheimer’s disease-affected areas based on analysis of segmented regions in MRI images. The Kaggle database information selected features and existing process observations from the first instance. The collected images are processed to enhance the image quality and accuracy for the clustering process, wherein the affected region is based on the segmentation process.

Figure 11. F-Measure Analysis

 

Error Rate

In Fig. 12, the detection of Alzheimer’s disease incorporates the portion of the brain cells analyzed using noisy pixel identification based on error detection. The metaheuristic optimized deep learning approaches in a consecutive manner considered error detection for performing additional clustering processes to improve the image quality. The noisy pixel does not identify, based on the condition Hr(Tf,Sf) and Hr(Tf) in a sequential manner of image segmentation. The histogram representation is used for analyzing the large data volume through machine learning techniques. The automatic AD detection from MRI image input is based on noisy pixels, preventing error detection. The causing errors can be addressed by detecting the biomarkers and ROI region through machine learning techniques to process accumulated data by applying the correlated information.

Figure 12. Error Rate Analysis

 

Computational Time

The study reveals that as the number of features increases, computational time generally increases across all algorithms that can be visualized in Fig. 13. DSC-NN and HCQMLM show the highest computational times, suggesting higher computational complexity or less optimized algorithms compared to EC+ML and FODC-CNN. FODC-CNN consistently outperforms other methods in terms of computational time, indicating its efficiency in processing data even as feature counts increase. These results underscore the importance of considering computational time when selecting algorithms for tasks like feature extraction and classification, especially in applications where processing speed is critical. Further optimization could improve performance and resource utilization in future studies.

Figure 13. Computational Time Analysis of the DSC-NN and Other Models

 

Discussion

The research shows that the FODC-CNN, an optimization-based densely linked convolutional neural network, could improve the diagnosis of AD by analyzing MRI images. The algorithm’s robustness in detecting regions impacted by AD in MRI scans was demonstrated by its excellent sensitivity, specificity, and accuracy rates. Analyses of the algorithm’s specificity and sensitivity show that it can detect small alterations that are signs of early-stage Alzheimer’s disease. Analysis using the F-measure demonstrates its capacity to optimize the trade-off between detecting genuine positives and avoiding false positives and false negatives, while also balancing recall and precision. Improving the accuracy of clustering segmentation is mostly due to the fly optimization algorithm. Nevertheless, the study does note certain limitations, including the fact that it only used one dataset and that its results may not apply to other groups. It is recommended that future studies enhance the algorithm, validate its applicability across different datasets, and think about doing longitudinal studies to see how well it tracks the evolution of diseases.
Factors such as the study scope, sample size, data gathering methods, statistical analysis, and regulatory issues dictate how long the study will take. Recruitment, data collecting, and analysis could take longer for larger research. Surveys and interviews are other methods that can alter the time required. Extending the timeframe of the study may be necessary for complex statistical analysis or the integration of data from different sources. Money and materials also play a role in how long it takes. Improving diagnostic methods and providing a dependable tool for early diagnosis are the goals of this work, which offers an automated way of diagnosing Alzheimer’s disease utilizing MRI image processing and deep learning algorithms. The study unveils an algorithm that could enhance the accuracy of MRI-based Alzheimer’s disease diagnoses. The algorithm can detect and intervene early because of its excellent sensitivity, specificity, and accuracy rates. It reduces subjectivity and unpredictability while providing an objective and repeatable diagnostic. Clinical settings can include the method to help analyze MRI images and improve decision-making. Possible directions for further study include developing better machine-learning algorithms, validating diagnostic models in a wider range of patient groups, and investigating new imaging biomarkers. This work adds to the growing body of literature calling for investigations into therapy response evaluation and longitudinal illness monitoring.

 

Conclusion

This article introduced a fly optimization-based densely connected convolution neural network method for identifying AD using MRI inputs. The textural and statistical features are extracted for which a modified histogram representation based on noisy pixels is identified. The biomarkers are updated using fly optimization for maximizing segmentation. The target and outputs are verified for the recurrent optimization steps based on the best solution. The clustering process is improved for the regions identified such that the processes are validated for new feature extractions. In the feature extraction process, the noisy pixel classification post-replacement is pursued to prevent detection errors. The proposed method performs all possible neural learning using the modified histogram output to improve sensitivity. The proposed method improves sensitivity, specificity, accuracy, and F-Measure by 15.52%, 15.62%, 9.01%, and 10.52%, respectively. This method reduces the error by 11.29%; this is observed for the varying features. According to the research, models that are trained on a particular data set might not be able to generalize well to other datasets or populations when put in different environments. The study might have overcome this shortcoming by using a more representative dataset and testing the model on other datasets or in actual clinical situations to see how well it performed. Algorithm performance, data accessibility, and clinical validation are some of the limitations of the fly optimization-based CNN technique for Alzheimer’s disease diagnosis. For use in actual healthcare settings, thorough validation is required, even though early trials indicate promise. Ethical and regulatory issues, clinical decision support systems, longitudinal analysis, and multi-modal integration should be the areas of future study. The accuracy of diagnoses can be enhanced through the integration of data from several imaging modalities and the development of clinical decision support systems.

 

Funding: The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Competing Interests: The authors have no relevant financial or non-financial interests to disclose.

Ethical standards: This article is original and contains unpublished material. The corresponding author confirms that all of the other authors have read and approved the manuscript and that no ethical issues are involved.

 

References

1. Shaikh, T. A., & Ali, R. (2019). Automated atrophy assessment for Alzheimer’s disease diagnosis from brain MRI images. Magnetic resonance imaging, 62, 167-173.
2. Feng, J., Zhang, S. W., Chen, L., & Alzheimer’s Disease Neuroimaging Initiative (ADNI. (2020). Identification of Alzheimer’s disease based on wavelet transformation energy feature of the structural MRI image and NN classifier. Artificial Intelligence in Medicine, 108, 101940.
3. Han, R., Liu, Z., & Chen, C. P. (2022). Multi-scale 3D convolution feature-based Broad Learning System for Alzheimer’s Disease diagnosis via MRI images. Applied Soft Computing, 120, 108660.
4. Li, Y., Luo, J., & Zhang, J. (2022). Classification of Alzheimer’s disease in MRI images using knowledge distillation framework: an investigation. International Journal of Computer Assisted Radiology and Surgery, 1-9.
5. Ge, C., Qu, Q., Gu, I. Y. H., & Jakola, A. S. (2019). Multi-stream multi-scale deep convolutional networks for Alzheimer’s disease detection using MR images. Neurocomputing, 350, 60-69.
6. Hao, X., Yao, X., Risacher, S. L., Saykin, A. J., Yu, J., Wang, H., … & Zhang, D. (2018). Identifying candidate genetic associations with MRI-derived AD-related ROI via tree-guided sparse learning. IEEE/ACM transactions on computational biology and bioinformatics, 16(6), 1986-1996.
7. Goenka, N., & Tiwari, S. (2022). AlzVNet: A volumetric convolutional neural network for multiclass classification of Alzheimer’s disease through multiple neuroimaging computational approaches. Biomedical Signal Processing and Control, 74, 103500.
8. Ushizima, D., Chen, Y., Alegro, M., Ovando, D., Eser, R., Lee, W., … & Grinberg, L. T. (2022). Deep learning for Alzheimer’s disease: Mapping large-scale histological tau protein for neuroimaging biomarker validation. NeuroImage, 248, 118790.
9. Wang, X., Huang, W., Su, L., Xing, Y., Jessen, F., Sun, Y., … & Han, Y. (2020). Neuroimaging advances regarding subjective cognitive decline in preclinical Alzheimer’s disease. Molecular Neurodegeneration, 15(1), 1-27.
10. Carmo, D., Silva, B., Yasuda, C., Rittner, L., Lotufo, R., & Alzheimer’s Disease Neuroimaging Initiative. (2021). Hippocampus segmentation on epilepsy and Alzheimer’s disease studies with multiple convolutional neural networks. Heliyon, 7(2), e06226.
11. Zhang, J., Zheng, B., Gao, A., Feng, X., Liang, D., & Long, X. (2021). A 3D densely connected convolution neural network with connection-wise attention mechanism for Alzheimer’s disease classification. Magnetic Resonance Imaging, 78, 119-126.
12. Li, F., Liu, M., & Alzheimer’s Disease Neuroimaging Initiative. (2019). A hybrid convolutional and recurrent neural network for hippocampus analysis in Alzheimer’s disease. Journal of neuroscience methods, 323, 108-118.
13. Yu, H., Yang, L. T., Zhang, Q., Armstrong, D., & Deen, M. J. (2021). Convolutional neural networks for medical image analysis: state-of-the-art, comparisons, improvement and perspectives. Neurocomputing, 444, 92-110.
14. Turkson, R. E., Qu, H., Mawuli, C. B., & Eghan, M. J. (2021). Classification of Alzheimer’s disease using deep convolutional spiking neural network. Neural Processing Letters, 53(4), 2649-2663.
15. Al-Adhaileh, M. H. (2022). Diagnosis and classification of Alzheimer’s disease using a convolution neural network algorithm. Soft Computing, 1-12.
16. Kim, M., Sekiya, H., Yao, G., Martin, N. B., Castanedes-Casey, M., Dickson, D. W., … & Koga, S. (2023). Diagnosis of Alzheimer disease and tauopathies on whole-slide histopathology images using a weakly supervised deep learning algorithm. Laboratory Investigation, 103(6), 100127.
17. Guo, H., & Zhang, Y. (2020). Resting state fMRI and improved deep learning algorithm for earlier detection of Alzheimer’s disease. IEEE Access, 8, 115383-115392.
18. Qiang, Y. R., Zhang, S. W., Li, J. N., Li, Y., Zhou, Q. Y., & Alzheimer’s Disease Neuroimaging Initiative. (2023). Diagnosis of Alzheimer’s disease by joining dual attention CNN and MLP based on structural MRIs, clinical and genetic data. Artificial Intelligence in Medicine, 145, 102678.
19. Alinsaif, S., Lang, J., & Alzheimer’s Disease Neuroimaging Initiative. (2021). 3D shearlet-based descriptors combined with deep features for the classification of Alzheimer’s disease based on MRI data. Computers in Biology and Medicine, 138, 104879.
20. Bai, T., Du, M., Zhang, L., Ren, L., Ruan, L., Yang, Y., … & Deen, M. J. (2022). A novel Alzheimer’s disease detection approach using GAN-based brain slice image enhancement. Neurocomputing, 492, 353-369.
21. Mehmood, A., Yang, S., Feng, Z., Wang, M., Ahmad, A. S., Khan, R., … & Yaqub, M. (2021). A transfer learning approach for early diagnosis of Alzheimer’s disease on MRI images. Neuroscience, 460, 43-52.
22. Zhang, J., He, X., Qing, L., Gao, F., & Wang, B. (2022). BPGAN: Brain PET synthesis from MRI using generative adversarial network for multi-modal Alzheimer’s disease diagnosis. Computer Methods and Programs in Biomedicine, 217, 106676.
23. Ebrahimi, A., Luo, S., Chiong, R., & Alzheimer’s Disease Neuroimaging Initiative. (2021). Deep sequence modelling for Alzheimer’s disease detection using MRI. Computers in Biology and Medicine, 134, 104537.
24. Feng, J., Zhang, S. W., Chen, L., Zuo, C., & Alzheimer’s Disease Neuroimaging Initiative. (2022). Detection of Alzheimer’s disease using features of brain region-of-interest-based individual network constructed with the sMRI image. Computerized Medical Imaging and Graphics, 98, 102057.
25. Guan, H., Wang, C., & Tao, D. (2021). MRI-based Alzheimer’s disease prediction via distilling the knowledge in multi-modal data. NeuroImage, 244, 118586.
26. de Souza, R. G., dos Santos Lucas e Silva, G., dos Santos, W. P., & de Lima, M. E. (2021). Computer-aided diagnosis of Alzheimer’s disease by MRI analysis and evolutionary computing. Research on Biomedical Engineering, 37(3), 455-483.
27. Jo, T., Nho, K., Risacher, S. L., & Saykin, A. J. (2020). Deep learning detection of informative features in tau PET for Alzheimer’s disease classification. BMC bioinformatics, 21(21), 1-13.
28. Ni, Y. C., Tseng, F. P., Pai, M. C., Hsiao, I. T., Lin, K. J., Lin, Z. K., … & Chuang, K. S. (2021). Detection of Alzheimer’s disease using ECD SPECT images by transfer learning from FDG PET. Annals of Nuclear Medicine, 35(8), 889-899.
29. Chen, X., Li, L., Sharma, A., Dhiman, G., & Vimal, S. (2022). The application of convolutional neural network model in diagnosis and nursing of MR imaging in Alzheimer’s disease. Interdisciplinary Sciences: Computational Life Sciences, 14(1), 34-44.
30. Helaly, H. A., Badawy, M., & Haikal, A. Y. (2022). Toward deep mri segmentation for alzheimer’s disease detection. Neural Computing and Applications, 34(2), 1047-1063.
31. Liu, Junxiu, Mingxing Li, Yuling Luo, Su Yang, Wei Li, and Yifei Bi. “Alzheimer’s disease detection using depthwise separable convolutional neural networks.” Computer Methods and Programs in Biomedicine 203 (2021): 106032.
32. Pan, Dan, Chao Zou, Huabin Rong, and An Zeng. “Early diagnosis of Alzheimer’s disease based on three-dimensional convolutional neural networks ensemble model combined with genetic algorithm.” Sheng wu yi xue Gong Cheng xue za zhi= Journal of Biomedical Engineering= Shengwu Yixue Gongchengxue Zazhi 38, no. 1 (2021): 47-55.
33. Shahwar, Tayyaba, Junaid Zafar, Ahmad Almogren, Haroon Zafar, Ateeq Ur Rehman, Muhammad Shafiq, and Habib Hamam. “Automated detection of Alzheimer’s via hybrid classical quantum neural networks.” Electronics 11, no. 5 (2022): 721.
34. Tanveer, Muhammad, Ashraf Haroon Rashid, M. A. Ganaie, Motahar Reza, Imran Razzak, and Kai-Lung Hua. “Classification of Alzheimer’s disease using ensemble of deep neural networks trained through transfer learning.” IEEE Journal of Biomedical and Health Informatics 26, no. 4 (2021): 1453-1463
35. Özçelik, Y. B., & Altan, A. (2023). Overcoming nonlinear dynamics in diabetic retinopathy classification: a robust AI-based model with chaotic swarm intelligence optimization and recurrent long short-term memory. Fractal and Fractional, 7(8), 598.
36. Yağ, İ., & Altan, A. (2022). Artificial intelligence-based robust hybrid algorithm design and implementation for real-time detection of plant diseases in agricultural environments. Biology, 11(12), 1732.
37. Özçelik, Y. B., & Altan, A. Classification Of Diabetic Retinopathy By Machine Learning Algorithm Using Entorpy-Based Features.
38. https://www.kaggle.com/datasets/tourist55/alzheimers-dataset-4-class-of-images?resource=download
39. https://www.kaggle.com/datasets/sachinkumar413/alzheimer-mri-dataset

© Serdi 2024