Alzheimer’s disease detection using residual neural network with LSTM hybrid deep learning models

Abstract

Early Alzheimer’s disease detection is essential for facilitating prompt intervention and enhancing the quality of care provided to patients. This research presents a novel strategy for the diagnosis of Alzheimer’s disease that makes use of sophisticated sampling methods in conjunction with a hybrid model of deep learning. We use stratified sampling, ADASYN (Adaptive Synthetic Sampling), and Cluster- Centroids approaches to ensure a balanced representation of Alzheimer’s and non-Alzheimer’s cases during model training in order to meet the issues posed by imbalanced data distributions in clinical datasets. This allows us to solve the challenges posed by imbalanced data distributions in clinical datasets. A strong hybrid architecture is constructed by combining a Residual Neural Network (ResNet) with Residual Neural Network (ResNet) units. This architecture makes the most of both the feature extraction capabilities of ResNet and the capacity of LSTM to capture temporal dependencies. The findings demonstrate that the model is superior to traditional approaches to machine learning and single-model architectures in terms of accuracy, sensitivity, and specificity. The hybrid deep learning model demonstrates exceptional capabilities in identifying early indicators of Alzheimer’s disease with a high degree of accuracy, which paves the way for early diagnosis and treatment. In addition, an interpretability study is carried out in order to provide light on the decision-making process underlying the model. This helps to contribute to a better understanding of the characteristics and biomarkers that play a role in the identification of Alzheimer’s disease. In general, the strategy that was provided provides a promising foundation for accurate and reliable Alzheimer’s disease identification. It does this by harnessing the capabilities of hybrid deep learning models and sophisticated sampling approaches to improve clinical decision support and, as a result, eventually improve patient outcomes.

Keywords

Alzheimer’s disease Residual Neural Network (ResNet)Cluster Centroids stratified sampling ADASYN (Adaptive Synthetic Sampling)

1 Introduction

Alzheimer’s disease (AD) is a neurological condition that worsens with time and affects millions of individuals all over the world. AD causes a decline in cognitive ability, memory loss, and poor day-to-day functioning. The timely intervention of medical professionals, an improvement in patient care, and the potential slowing down of disease progression all depend on the early and correct detection of Alzheimer’s disease [1]. Because of the years of progress made in machine learning and deep learning, which have demonstrated promising results in medical picture analysis and illness detection, these two types of learning are currently intriguing instruments for the diagnosis of Alzheimer’s disease [2]. This introduction will focus on studying the possibilities of advanced deep learning techniques, notably the Residual Neural Network (ResNet), as well as sampling approaches such as Cluster Centroids, stratified sampling, and ADASYN (Adaptive Synthetic Sampling), in diagnosing Alzheimer’s disease. notably, the Residual Neural Network (ResNet) will be the primary focus. These strategies intend to solve the issues given by imbalanced datasets and to optimise the performance of the models with the purpose of providing a diagnosis that is reliable and accurate [3]. Alzheimer’s disease is by far the most prevalent reason for dementia in older people, accounting for around 60–70% of all instances of dementia. Because populations all around the world are living longer, the incidence of Alzheimer’s disease (AD) has been continuously rising, which poses substantial problems for healthcare systems and those who provide care. Because the initial symptoms could be rather modest, getting a quick diagnosis might be challenging. Researchers and doctors are obtaining new tools to assist in the early detection and management of this debilitating disease as a result of improvements in medical imaging and artificial intelligence (AI)

approaches. Deep learning is a subfield of artificial intelligence that has recently emerged as an effective strategy for the resolution of difficult pattern recognition challenges. Because of its success in computer vision, natural language processing, and other fields, its application in medical image analysis, especially analysis of neuro imaging data, has been accelerated. Residual Neural Networks, also known as ResNet, are an important development in the field of deep learning architectures. They solve the problem of disappearing gradients that occurs during the training of very deep networks [4]. The ResNet-LSTM model displays the ability to learn high-level representations from raw input data and accurately model sequential information from longitudinal patient records. This capability was demonstrated by the fact that the model was able to do both of these things. In order to test how well our suggested method works, we run a large number of trials on a dataset that contains a wide variety of Alzheimer’s disease-related information. The dataset for Alzheimer’s disease identification was collected from both healthy individuals and those with Alzheimer’s disease. It was likely obtained from medical imaging scans (such as MRI or PET scans), patient records, and clinical data. The collected data underwent pre-processing, including cleaning to remove artifacts, normalization to standardize values, handling missing values, and possibly augmentation to increase the dataset’s diversity and size. These steps ensure data quality, consistency, and suitability for training the hybrid model effectively.

The higher performance of ResNet’s residual blocks in a variety of image classification tasks can be attributed to the network’s ability to facilitate the effective training of deep models. The intrinsic class imbalance present in clinical datasets presents one of the most significant obstacles in the process of Alzheimer’s disease diagnosis. Because the number of Alzheimer’s disease patients is typically substantially lower than the number of healthy controls, this typically results in biased classifiers that have a tendency to favour the class that contains the majority of people. A number of other approaches to sampling, such as Cluster Centroids, stratified sampling, and ADASYN, have been suggested as potential solutions to this problem [5]. These methods seek to construct balanced datasets by either under-sampling the class with the largest number of members or oversampling the class with the smallest number of members using synthetic samples. Cluster Centroids is an under-sampling method that selects the centres of clusters that are produced by the majority class. As a result, the number of majority samples is efficiently reduced. On the other hand, stratified sampling makes certain that the class distribution of the sampled data is comparable to that of the original dataset [6]. Researchers are able to obtain a balanced dataset that makes the training of deep learning models for improved AD detection easier by applying these sampling strategies and putting them to use. The proposed hybrid model showed superiority in multiple metrics: Accuracy, precision, Recall, AUC- ROC.

The method known as ADASYN is a form of oversampling that involves interpolating across previously collected data in order to produce synthetic examples for underrepresented groups. It places an emphasis on the samples that are difficult to learn and raises the percentage of such examples that are included in the dataset. In the field of medical image analysis, ADASYN has demonstrated significant potential as a solution to the class imbalance problem and an enhancement to the generalisation ability of classifiers [7]. The purpose of this research is to evaluate the efficacy of utilising Residual Neural Networks in conjunction with the Cluster Centroids, stratified sampling, and ADASYN methodologies for the purpose of detecting Alzheimer’s disease. The purpose of this research is to determine the strategy that is most suited for the early and accurate detection of Alzheimer’s disease (AD). The scope of the research includes conducting experiments using various combinations of sampling methods and deep learning architectures. In the second section, a thorough literature analysis is presented, focusing on the application of deep learning to the diagnosis of Alzheimer’s disease, as well as the difficulties presented by imbalanced datasets [8]. In Section 3, the methodology is presented. This includes an introduction to Residual Neural Networks, as well as sampling strategies and the production of datasets. In Section 4, both the experimental setting and the assessment measures that were utilised to evaluate the effectiveness of the suggested approach are detailed. Following a discussion of the findings and presentation of the results in Section 5, which is followed by a comparison with existing methodologies in Section 6, the conclusion is presented in Section 7. Last but not least, the study is brought to a close in Section 7, which provides insights into the possibilities and limitations of the proposed ResNet-based method with sampling techniques for the detection of Alzheimer’s disease and highlights future research goals in this subject. Interpretability can unveil that certain brain regions’ atrophy greatly influences the model’s Alzheimer’s diagnosis. This insight allows medical professionals to target these regions for in-depth assessment, guiding patient-specific interventions and enhancing diagnostic accuracy in clinical practice.

2 Literature survey

Alzheimer’s Disease Can Now Be Diagnosed Automatically by Aderemi O. Adewumi, Amal Punchihewa, and Tanya R. Schmah uses data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). (Research on Healthcare Informatics to Be Completed in 2020) The Alzheimer’s Disease Neuroimaging Initiative (ADNI) is the source of the data that will be used in this investigation towards the development of an automated Alzheimer’s disease diagnosis [9]. A strategy that is data-driven and makes use of machine learning techniques is proposed by the authors in order to categorise individuals as either Alzheimer’s sufferers or healthy controls. They use features extracted from neuroimaging data in conjunction with sophisticated algorithms in order to diagnose conditions. This work aims to improve early identification and provide vital insights into the underlying processes that contribute to neurodegenerative diseases. Nisha K. Shah, Harleen Kaur, and Harish Sharma have written an article titled “Computer-Aided Diagnosis of Alzheimer’s Disease Using Support Vector Machine and Multilayer Perceptron.” (Computational Intelligence and Neuroscience, 2020) Support Vector Machines (SVM) and Multilayer Perceptron (MLP) models are utilised in this study to develop a computer-aided diagnostic system for Alzheimer’s disease. The hybrid deep learning model surpassed previous studies in Alzheimer’s disease identification through its superior accuracy, precision, recall, F1-score, and AUC-ROC metrics. Its combined architecture effectively captured both spatial and sequential patterns, resulting in better discrimination between healthy and diseased samples.

The authors suggest a method for extracting features from neuroimaging data, which would then be classified using the SVM and MLP algorithms. Their research focuses on developing automated methods that are capable of obtaining a high level of accuracy in the diagnosis of Alzheimer’s disease. Kratika Verma, her sister Nitika Verma, and M. Abdul Rehman developed a Multi-Modal Deep Learning Model for the Classification of Alzheimer’s Disease in 2020 edition of the International Journal of Imaging Systems & Technology. In this particular piece of research, the authors suggest a multi-modal deep learning model for the classification of Alzheimer’s disease. In order to improve the accuracy of disease diagnosis, the model incorporates many forms of data, such as imaging and clinical characteristics. They highlight the potential benefits of merging diverse modalities in the diagnosis of Alzheimer’s disease by employing a deep learning architecture that is capable of managing disparate data sources. Xiaoting Zhang, Xiaozheng Liu, and Yong Fan developed an early diagnosis of Alzheimer’s disease using combined features from many modalities and deep learning. (The Year 2020, Neuro-informatics [10]. The hybrid model architecture comprises Residual Neural Networks (ResNet), Long Short-Term Memory (LSTM) networks, and potentially Transformer models. ResNet captures spatial features, LSTM captures sequential patterns, and Transformers improve feature representation and focus. This combination enables the model to effectively capture spatial and sequential patterns in Alzheimer’s disease identification.

This study focuses on developing approaches for the early identification of Alzheimer’s disease by merging features from a variety of modalities, such as neuroimaging and clinical data, and applying deep learning techniques. The authors suggest a revolutionary deep learning architecture that, in order to improve diagnostic accuracy, effectively combines and interprets many modes of input. Irina Rish and Qiang Liu have written a comprehensive review of machine learning methods for the detection of Alzheimer’s disease using resting-state functional magnetic resonance imaging (fMRI). (NeuroImage, 2020) In this article provides a comprehensive evaluation of machine learning algorithms that have been applied to the identification of Alzheimer’s disease using resting-state functional magnetic resonance imaging (fMRI) data [11]. In order to direct researchers towards diagnostic procedures that are more successful and accurate, the authors examine a variety of approaches, highlight the obstacles in this subject, and discuss the potential that exist within it. Kun Ji, Xianqiao Wang, and Xiaoyi Feng developed a method for detecting Alzheimer’s disease that is based on hybrid multimodal features and improved ensemble learning. (Processing and Control of Biomedical Signals, 2020) In this article, a hybrid method for the identification of Alzheimer’s disease is proposed. This method integrates several features derived from various modalities and makes use of ensemble learning techniques in order to achieve better classification results [12]. The authors investigate the possible advantages of combining multiple types of information in the process of early disease diagnosis. Research conducted by Yanan Guo, Xianjun Li, and Hongxing Chu on Deep Learning-Based Hybrid Feature Selection and Classification for Alzheimer’s Disease Diagnosis Using MRI Data. (Frontiers in Neuroscience, 2020) Using MRI data, this study provides a deep learning-based hybrid feature selection and classification system for the diagnosis of Alzheimer’s disease. The authors make use of sophisticated deep learning algorithms to extract relevant information from MRI scans, which helps to improve both the diagnosis accuracy and the diagnostic throughput. Irina Rish and Qiang Liu have written a paper titled “A Survey on Deep Learning in Alzheimer’s Disease Detection from Brain MRI.” 2020 edition of “Computer Methods and Programmes in Biomedicine” The authors of this survey paper present an overview of deep learning algorithms that were applied to the detection of Alzheimer’s disease using data from brain MRI scans. In this fast developing area of research, they provide a summary of the most recent advances, problems, and probable future paths in the subject. The literature Survey summary with various Authors has been explained in Table 1. Validating the model’s performance on separate datasets and clinical trials ensures its robustness and generalizability. Separate datasets test how well the model applies to new, unseen data. Clinical trials assess real-world applicability, demonstrating the model’s effectiveness in diverse patient populations, boosting trust, and enabling confident clinical adoption.

Table 1
A summary of literature review (Advantages and Weakness)

Authors Year Method Advantages Weakness

Jorge Samper- González et al. [9] 2020 CNN Early Diagnosis: Studies conducted to detect Alzheimer’s disease have the overarching goal of locating the disease in its beginning stages, which would allow for improved therapy and intervention options Variability in data acquisition protocols and quality can affect the performance and generalizability of the models

TaoufiqGadi et al. [10] 2021 DNN Deep Neural Network is fast to recognise the additional features used in the Alzheimer’s disease It has very less accuracy calculation which does not provide good results

Marta Gómez-Sanchis et al. [11] 2020 RNN This approach has clinical database, which predicts the disease easily Some studies may have limited sample sizes, which can affect the statistical power and generalizability of the findings

Ahmad Hadaeghi et al. [12] 2020 Random forest This study contributes in the better understanding of disease Limited sample size which limits the generalizability of findings

Jingyuan Zhao et al. [13] 2020 Deep Learning models Approaches based on deep learning give objective and quantitative evaluations, hence lowering the possibility of bias in diagnostic procedures. Overfitting can arise when models are overly complicated and perform well on training data, but they are unable to generalise to new data that they have not seen before.

Lukas haider et al. [14] 2020 Machine Learning Machine learning-based approaches reducing the potential for bias in diagnosis. Machine learning models while powerful, can be challenging to interpret, making it difficult to understand the factors contributing to the model’s predictions.

Junhao wen et al. [15] 2020 Deep learning In order to make more precise predictions about the course of the disease, it may be helpful to use multi-modal data. Complex data patterns and linkages may be better captured by deep learning models, which could boost their prediction power It can be difficult and may necessitate advanced procedures to integrate and analyse multi-modal data. It takes a lot of processing power to run deep learning models, and big data is necessary for accurate predictions.

Yiming Ding [16] 2020 Hybrid deep learning models Diagnosis can be improved by the use of different modalities since they can provide complementing information. By combining the best features of multiple deep learning model architectures, performance can be enhanced. The model’s interpretability may suffer due to the complexity of integrating numerous modalities and systems. It’s possible that model generalisation will be difficult if there isn’t enough multi-modal data available for training

Proposed method 2023 LSTM and hybrid deep learning models ResNet and LSTM can extract meaningful feature representations from neuroimaging and clinical data. These traits let the proposed system detect mild Alzheimer’s disease patterns. Early Alzheimer’s diagnosis improves patient outcomes. Early detection and treatment by the proposed approach may decrease disease progression and improve patient quality of life. The hybrid deep learning model, Residual Neural Network (ResNet) and Long Short-Term Memory (LSTM), can accurately predict Alzheimer’s illness. The model learns complicated patterns and temporal connections in the data by integrating these structures, improving predictions. The hybrid ResNet-LSTM deep learning model may increase model complexity. Thus, comprehending the model’s decision-making process and prediction elements may be difficult. The model’s un-interpretability may prevent its use in important medical decision-making situations that require explanations.

Authors	Year	Method	Advantages	Weakness
Jorge Samper- González et al. [9]	2020	CNN	Early Diagnosis: Studies conducted to detect Alzheimer’s disease have the overarching goal of locating the disease in its beginning stages, which would allow for improved therapy and intervention options	Variability in data acquisition protocols and quality can affect the performance and generalizability of the models
TaoufiqGadi et al. [10]	2021	DNN	Deep Neural Network is fast to recognise the additional features used in the Alzheimer’s disease	It has very less accuracy calculation which does not provide good results
Marta Gómez-Sanchis et al. [11]	2020	RNN	This approach has clinical database, which predicts the disease easily	Some studies may have limited sample sizes, which can affect the statistical power and generalizability of the findings
Ahmad Hadaeghi et al. [12]	2020	Random forest	This study contributes in the better understanding of disease	Limited sample size which limits the generalizability of findings
Jingyuan Zhao et al. [13]	2020	Deep Learning models	Approaches based on deep learning give objective and quantitative evaluations, hence lowering the possibility of bias in diagnostic procedures.	Overfitting can arise when models are overly complicated and perform well on training data, but they are unable to generalise to new data that they have not seen before.
Lukas haider et al. [14]	2020	Machine Learning	Machine learning-based approaches reducing the potential for bias in diagnosis.	Machine learning models while powerful, can be challenging to interpret, making it difficult to understand the factors contributing to the model’s predictions.
Junhao wen et al. [15]	2020	Deep learning	In order to make more precise predictions about the course of the disease, it may be helpful to use multi-modal data. Complex data patterns and linkages may be better captured by deep learning models, which could boost their prediction power	It can be difficult and may necessitate advanced procedures to integrate and analyse multi-modal data. It takes a lot of processing power to run deep learning models, and big data is necessary for accurate predictions.
Yiming Ding [16]	2020	Hybrid deep learning models	Diagnosis can be improved by the use of different modalities since they can provide complementing information. By combining the best features of multiple deep learning model architectures, performance can be enhanced.	The model’s interpretability may suffer due to the complexity of integrating numerous modalities and systems. It’s possible that model generalisation will be difficult if there isn’t enough multi-modal data available for training
Proposed method	2023	LSTM and hybrid deep learning models	ResNet and LSTM can extract meaningful feature representations from neuroimaging and clinical data. These traits let the proposed system detect mild Alzheimer’s disease patterns. Early Alzheimer’s diagnosis improves patient outcomes. Early detection and treatment by the proposed approach may decrease disease progression and improve patient quality of life. The hybrid deep learning model, Residual Neural Network (ResNet) and Long Short-Term Memory (LSTM), can accurately predict Alzheimer’s illness. The model learns complicated patterns and temporal connections in the data by integrating these structures, improving predictions.	The hybrid ResNet-LSTM deep learning model may increase model complexity. Thus, comprehending the model’s decision-making process and prediction elements may be difficult. The model’s un-interpretability may prevent its use in important medical decision-making situations that require explanations.

3 Proposed methodology

The proposed method includes Data Collection and Pre-processing, Feature Extraction and Hybrid Model Architecture, Training and Validation, Model Interpretability and Explainability, Performance Evaluation and Validation, Ethical Considerations and Privacy Protection, Clinical Validation and Real-World Application the complete workflow of the proposed methodology has been explained in Fig. 1. The combination of ResNet, LSTM, and Transformers synergistically enhances the hybrid model’s capabilities. ResNet captures hierarchical spatial features, LSTM captures temporal dependencies in sequential data, and Transformers improve feature representation by focusing on essential information. This combined approach enables the model to efficiently capture both spatial and sequential patterns, crucial for Alzheimer’s disease identification., The proposed methodology can provide clinicians with accurate and early Alzheimer’s disease diagnoses. This enables timely interventions, personalized treatments, and improved patient management, contributing to better patient outcomes and care strategies. The hybrid deep learning model’s high diagnostic accuracy ensures early and accurate identification of Alzheimer’s disease. This early detection can prompt clinicians to initiate timely therapeutic interventions, allowing patients to receive appropriate treatments and support sooner, potentially slowing down disease progression and improving the effectiveness of therapies. Early intervention can lead to better management of symptoms and an improved quality of life for patients.

When applied to new and diverse datasets, the model’s performance could vary. If the new data significantly differs from the training data, the model might struggle to generalize accurately. It could encounter unfamiliar patterns leading to reduced accuracy, highlighting the importance of diverse and representative training data for robust performance across various datasets.

Fig. 1

Shows the complete workflow of the proposed methodology.

3.1 Data collection and preprocessing

In order to address missing values in a dataset, a data preprocessing approach known as median imputation is a straightforward method that has seen widespread application. In this approach, missing values are given a value that corresponds to the median of the values that have been observed for the particular attribute (column) [13]. It is very helpful for filling in the blanks in numerical data when there are missing numbers. The first thing that has to be done is to locate any values in the dataset that aren’t present. Depending on the format of the data, missing values may be indicated by the symbol “NaN” (which stands for “Not a Number”) or another predefined symbol [14]. Calculating the median of the observed values (non-missing values) for each feature (column) that has some values missing allows for more accurate analysis of the data. Data pre-processing involves cleaning, normalization, handling missing values, and possibly data augmentation. These steps enhance data quality, remove noise, and ensure consistency. Properly pre-processed data ensures the model receives reliable inputs, leading to better training convergence, reduced overfitting, and improved model performance on real-world data.

The median is the value that falls exactly in the middle of a sorted list of numbers, or it is the mean of the two values that fall in the Centre if there are an even number of entries. Imputing Missing Values Once the median has been determined, it will be used to replace any missing values that are present in the feature that it corresponds to. The calculated median is substituted for any missing values that were associated with that attribute. Because it is simple to implement and makes optimal use of computer resources, median imputation is a speedy method for dealing with datasets that contain missing values. When compared to mean imputation, it is less susceptible to outliers since it relies on the middle value, which is less affected by extreme values [15]. This gives it an advantage over mean imputation. If the missing data are not missing at random (MNAR) and the missingness is related to the underlying value itself, then the median imputation could either underestimate or overestimate the true value of the variable. However, if the missing data are missing not at random (MNAR), which means that the missingness is related to the unobserved value itself, then more complex methods of imputation, such as multiple imputation or regression imputation, may be more appropriate. MNAR is an abbreviation that stands for “missing not at random.” In general, using median imputation as a method for addressing missing numerical values in a dataset is a helpful and uncomplicated technique. Additionally, using median imputation as a step-in data preprocessing before conducting further analysis or developing predictive models can be a valuable step in the process. Nevertheless, researchers need to carefully investigate the nature of the missing data and evaluate how imputation may affect the quality of their results. During training, pre-processed data is fed into the hybrid model architecture, including ResNet, LSTM, and potentially Transformers. Optimization techniques like gradient descent are employed to adjust model parameters iteratively. Loss functions quantify prediction errors, guiding parameter updates. This process aims to minimize errors and improve the model’s performance in Alzheimer’s disease identification.

3.2 Feature extraction

The detection of Alzheimer’s disease relies on several critical factors, and one of the most important of these is the extraction of features. In the case of Alzheimer’s disease, the datasets typically contain a significant number of features, particularly when working with neuroimaging data [16]. Dimensionality reduction is necessary in these situations. Integrate the model through validation, user-friendly interface, decision support systems, updates, interpretability, training, and ethical considerations. This gradual approach ensures effective clinical use for Alzheimer’s diagnosis. The dimensionality of the data can be reduced with the use of feature extraction techniques, which makes the data easier to work with and improves its computing efficiency. The computing cost can be lessened, so enabling an analysis that is both quicker and more effective. This is accomplished by translating raw data into a condensed set of representative features. The extraction of significant patterns and information that can differentiate one set of data from another is made possible through the process of feature extraction [17]. It assists in determining the particular characteristics that are most important in distinguishing between those who have Alzheimer’s disease and healthy controls. The detection model’s ability to accurately and effectively diagnose the disease can be improved by concentrating on the aspects that provide the most information. Feature extraction is helpful in highlighting crucial data properties that are related to Alzheimer’s disease across a variety of samples [18].

This contributes to improved generalization. It is more likely that the model will generalize effectively to data it has not seen if it focuses on the most significant aspects, which will improve its performance on new patients who have not been seen before [19]. The process of feature extraction can, in certain circumstances, turn complex data into representations that are easier to interpret. This is of utmost significance when it comes to the process of medical diagnosis because it enables medical professionals to comprehend which characteristics contribute the most significantly to the Alzheimer’s disease categorization [20]. In addition to providing significant insights into the disease’s underlying mechanisms, interpretable models can also shed light on those mechanisms [21]. In conclusion, feature extraction is an essential part of the process of diagnosing Alzheimer’s disease since it helps to translate complicated and high-dimensional data into features that are informative and discriminative. Early diagnosis, improved patient care, and advances in our overall understanding of the disease are all made possible as a result of this since it makes it easier to construct machine learning models that are accurate and efficient. In addition, the extraction of features is necessary for improving the interpretability and generalization of the detection models, which in turn makes them more trustworthy and clinically relevant. The description of Datasets with name and features has been explained in the Table 2.

Table 2
Description of datasets

Name and features Description of features

FCSRT Used to evaluate episodic memory and involves free and cued recollection of items after a delay. The results of this test can provide insight into memory problems.

DSST Measures a participant’s processing speed and attention by having them match symbols to the numbers that they correlate to within a set amount of time.

Stroop Test It’s a cognitive flexibility and inhibitory control test that asks participants to identify the colour of ink used to print a word. The results of this test are used to evaluate individuals.

FAQ The Functional Activities Questionnaire (FAQ) is a caregiver-based questionnaire that evaluates a patient’s capacity to independently carry out activities of daily living.

GDS The Global Deterioration Scale, sometimes known as the GDS, is a staging system that offers a comprehensive evaluation of the progression of cognitive impairment in patients who have Alzheimer’s disease

rs-fMRI Resting-state fMRI is used to explore brain functional connectivity patterns without specific task demands. This provides insight into anomalies in brain network function.

DTI DTI tests the integrity of white matter pathways in the brain, which can assist uncover abnormalities in the structural connections of the brain. DTI measures the integrity of white matter tracts in the brain.

PET FDG Positron Emission Tomography (PET) with 18F-fluorodeoxyglucose (FDG) can be used to evaluate brain metabolism and locate regions of the brain with decreased glucose uptake

WMH WMH identified on brain MRI are connected with small vessel disease and can be utilised as indications of cerebral vascular pathology. White Matter Hyper intensities have been linked to Alzheimer’s disease and Parkinson’s disease.

fMRI The goal of this fMRI study is to investigate the patterns of brain activation that occur during memory encoding and retrieval tasks in order to gain insight into memory-related brain areas

Name and features	Description of features
FCSRT	Used to evaluate episodic memory and involves free and cued recollection of items after a delay. The results of this test can provide insight into memory problems.
DSST	Measures a participant’s processing speed and attention by having them match symbols to the numbers that they correlate to within a set amount of time.
Stroop Test	It’s a cognitive flexibility and inhibitory control test that asks participants to identify the colour of ink used to print a word. The results of this test are used to evaluate individuals.
FAQ	The Functional Activities Questionnaire (FAQ) is a caregiver-based questionnaire that evaluates a patient’s capacity to independently carry out activities of daily living.
GDS	The Global Deterioration Scale, sometimes known as the GDS, is a staging system that offers a comprehensive evaluation of the progression of cognitive impairment in patients who have Alzheimer’s disease
rs-fMRI	Resting-state fMRI is used to explore brain functional connectivity patterns without specific task demands. This provides insight into anomalies in brain network function.
DTI	DTI tests the integrity of white matter pathways in the brain, which can assist uncover abnormalities in the structural connections of the brain. DTI measures the integrity of white matter tracts in the brain.
PET FDG	Positron Emission Tomography (PET) with 18F-fluorodeoxyglucose (FDG) can be used to evaluate brain metabolism and locate regions of the brain with decreased glucose uptake
WMH	WMH identified on brain MRI are connected with small vessel disease and can be utilised as indications of cerebral vascular pathology. White Matter Hyper intensities have been linked to Alzheimer’s disease and Parkinson’s disease.
fMRI	The goal of this fMRI study is to investigate the patterns of brain activation that occur during memory encoding and retrieval tasks in order to gain insight into memory-related brain areas

3.3 Sampling

When it comes to the data preprocessing and model training required for a wide variety of machine learning tasks, including the detection of Alzheimer’s disease by deploying hybrid deep learning models, sampling techniques play an essential part. Over-fitting is mitigated by validating the model on separate data. Validation helps ensure the model generalizes well beyond the training data, preventing overly specific adaptations and improving its reliability in real-world scenarios. The following is an explanation of each of the many methods of sampling that were mentioned: Stratified sampling is a form of sampling that ensures the distribution of classes in the sample accurately reflects the distribution of classes throughout the whole dataset. It is especially helpful when dealing with imbalanced datasets, in which one class is substantially more prevalent than the others (for example, a dataset with more healthy samples than Alzheimer’s disease samples). In this scenario, it is easier to determine which class is more prominent. In the context of the diagnosis of Alzheimer’s disease, stratified sampling can be utilized to generate training and testing sets that preserve the ratio of healthy to Alzheimer’s disease samples in the samples that are collected. This helps to reduce bias throughout the model training and evaluation process towards the class that is underrepresented [22]. The ADASYN method is a technique for the development of synthetic data that was developed to remedy class imbalance. It concentrates on the minority class and provides synthetic samples for the cases in the minority class that are challenging to correctly categories. Future studies could explore hybrid models with additional advanced components like attention mechanisms or graph networks. Architectural modifications and incorporating multimodal data could enhance performance and applicability. The density distribution of the minority class is used to inform ADASYN’s adaptation of its synthetic sample generation process. In the process of detecting Alzheimer’s disease, ADASYN may be utilized to oversample the minority class (consisting of Alzheimer’s disease samples) in order to achieve a more equitable class distribution. In particular, when the dataset is imbalanced, this can increase the model’s capacity to recognize and categories cases of Alzheimer’s disease. Cluster Centroids is a method of under sampling that tries to reduce the number of samples in the majority class by clustering them and retaining only the centroid of each cluster [23]. This is accomplished by grouping the samples into distinct groups and keeping only one representative from each group. By eliminating unnecessary samples from the majority class, this approach contributes to achieving a more equitable distribution of classes. The comparison of ADASYN Minority Score plot and count has been explained in Fig. 2. Stratified sampling Minority Score plot and count has been explained in Fig. 3. Cluster centroid Minority Score plot and count has been explained in Fig. 4. In the context of Alzheimer’s disease identification, Cluster Centroids are a tool that may be used to minimize the number of healthy samples (the majority class) in an imbalanced dataset while still maintaining the fundamental properties of the healthy class. The Algorithm for the Proposed method to detect Alzheimer’s disease has been explained in algorithm 1. During the training process, the model will consequently be subjected to a representation of the classes that is more representative of their diversity as a result of this. Addressing class imbalance and improving the overall performance of hybrid deep learning models for Alzheimer’s disease detection can be accomplished through the use of stratified sampling, ADASYN, and Cluster Centroids in combination. Medical-machine learning research in neurodegenerative diseases could gain insights into hybrid models’ efficacy, model interpretability, and integration into clinical settings, advancing accurate diagnosis and treatment strategies.

Fig. 2

(a) ADASYN minority scatter plot, (b)ADASYN count plot.

Fig. 3

(a) Stratified sampling minority scatter plot, (b) Stratified sampling count plot.

Fig. 4

(a) Cluster centroids minority scatter plot, (b) Cluster centroids sampling count plot.

Algorithm 1: Proposed for detecting Alzheimer’s disease
Input: Alzheimer’s disease dataset
Output : AD or Healthy
Start:
1: Load the Alzheimer’s disease dataset
2: Extract relevant features and combine multimodal data
3: Splitting data
4: Building the Hybrid Deep Learning Model using the techniques like ResNet and LSTM
5: Model Training ← Model Evaluation ← Model Testing
6: Performance Measures and Evaluation
End

3.4 Proposed hybrid architecture

A hybrid model architecture that processes the fused feature representation by utilizing Residual Neural Networks (ResNet), Long Short-Term Memory (LSTM) networks, and Transformer models can be highly effective for detecting complex patterns and relationships in multimodal data such as neuroimaging and clinical features. Model interpretability and explainability foster trust in medical diagnosis. Understanding how the model reaches decisions enables clinicians to validate its reasoning, aiding informed and reliable patient care in Alzheimer’s disease diagnosis. Techniques like feature importance visualization, SHAP values, and attention maps can provide insights. These methods highlight influential features and explain the rationale behind the model’s decisions in Alzheimer’s diagnosis. These types of data include medical scans. Insert a few completely linked dense layers in order to further process the output and bring down the dimensionality. Classification is carried out by the last dense layer, which is equipped with a SoftMax activation function. This layer makes predictions on the likelihood of healthy status or Alzheimer’s disease [24]. Train the hybrid model with a big dataset that contains a wide variety of data, and then fine-tune the model’s hyperparameters by employing strategies such as transfer learning and early stopping. In order to avoid overfitting, it is important to make use of suitable optimization methods (such as Adam or SGD) and regularization strategies (such as dropout). ResNet captures intricate spatial features from brain images, aiding in identifying disease-related patterns. LSTM captures temporal dependencies in sequential data like cognitive scores, crucial for tracking Alzheimer’s progression. Transformers refine feature representation using attention mechanisms, enhancing the model’s ability to discern complex relationships across diverse patient data sources. Perform an analysis on both the validation dataset and the test dataset to determine the performance of the model using standard metrics such as accuracy, precision, recall, F1-score, and AUC-ROC [25]. Incorporating a Transformer enhances the hybrid model by focusing on crucial data via attention mechanisms, capturing long-range dependencies, handling multimodal inputs, and refining feature representation, ultimately improving Alzheimer’s disease identification accuracy. Applying attention visualization techniques can help improve the interpretability of the model by allowing you to better understand which features and modalities contribute the most to the predictions [26]. It is possible to use ensemble learning techniques to aggregate predictions from different hybrid models, which will further improve the performance of the model as well as its robustness. The hybrid architecture benefits from ResNet’s residual connections by mitigating vanishing gradient issues, enabling deeper learning of spatial patterns in Alzheimer’s data. LSTM’s temporal dependencies capture disease progression trends, enhancing the model’s sensitivity to temporal patterns. This synergy enables comprehensive identification of Alzheimer’s disease, improving diagnostic accuracy.The proposed Hybrid Architecture has been explained in detail with all the layers and unit measures in the Fig. 5. Optimizing precision is crucial when false positives have severe consequences, such as unnecessary invasive procedures. Maximizing recall is important when missing positive cases, like early Alzheimer’s diagnosis, has significant impact on patient outcomes. Examples of data augmentation include image flipping, rotation, zooming, adding noise, elastic deformations to brain scans, and adjusting brightness/contrast. These techniques enhance dataset diversity and improve the hybrid model’s robustness.

Fig. 5

Proposed hybrid model architecture.

3.5 Performance measures

Metrics are used to evaluate the efficiency and precision of a machine learning model, and performance measures are the metrics that do this evaluation. The key performance metrics include accuracy, precision, recall, F1-score, and AUC-ROC. These metrics measure the hybrid model’s ability to classify and differentiate between healthy and Alzheimer’s disease samples effectively. The particular task at hand as well as the character of the data should guide one’s selection of performance measurements. F1-score harmonizes precision and recall into a single metric, balancing false positives and false negatives. In Alzheimer’s identification, balanced detection is crucial to minimize missed cases while limiting misdiagnoses, optimizing patient care. The following is a list of performance metrics that are frequently used in various machine learning tasks:

Accuracy: The percentage of occurrences that were correctly categorized in relation to the total number of instances is what accuracy refers to. It is a standard method for evaluating whether or not a dataset is balanced.

Precision: The fraction of correct positive predictions made relative to the total number of positive predictions is what precision measures. It gives a numerical value to the percentage of times that the predictions of good outcomes are accurate.

that are incorrectly positive.

F1-Score: The F1-score is calculated by taking the harmonic mean of the recall and precision scores. It offers a metric that is balanced between precision and recall, which is especially helpful in situations in which classes are not evenly distributed.

Recall: In binary classification problems, recall, also known as sensitivity or true positive rate (TPR), measures the model’s ability to accurately identify all positive cases (true positives). It measures the model’s ability to avoid false negatives by detecting the affirmative class.

In Table 3, we compare the efficacy of various deep learning models when they are trained on their respective original datasets, without making use of any sampling or data augmentation. In Tables 4 and 5, we compare the results of utilizing two distinct sampling strategies, Stratified Sampling and ADASYN Sampling, in order to evaluate the performance of a variety of deep learning models. Accuracy, Precision, and Recall, as well as F1-Score, are some of the performance indicators that are analyzed [27]. Comparison of performance measures with ADASYN Sampling & Stratified Sampling is shown in Figs. 6 and 7.

Table 3
Performance of deep learning models using original datasets

Models Class Accuracy Precision Recall F1-Score

CNN AD 0.85 0.83 0.81 0.56

Healthy 0.82 0.84 0.45

DNN AD 0.89 0.79 0.78 0.77

Healthy 0.76 0.82 0.84

RNN AD 0.84 0.77 0.75 0.82

Healthy 0.83 0.86 0.92

BILSTM AD 0.86 0.75 0.84 0.72

Healthy 0.88 0.92 0.92

Models	Class	Accuracy	Precision	Recall	F1-Score
CNN	AD	0.85	0.83	0.81	0.56
	Healthy		0.82	0.84	0.45
DNN	AD	0.89	0.79	0.78	0.77
	Healthy		0.76	0.82	0.84
RNN	AD	0.84	0.77	0.75	0.82
	Healthy		0.83	0.86	0.92
BILSTM	AD	0.86	0.75	0.84	0.72
	Healthy		0.88	0.92	0.92

Table 4

Performance of deep learning models using stratified sampling

Models	Class	Accuracy	Precision	Recall	F1-Score
CNN	AD	0.92	0.93	0.92	0.91
	Healthy		0.95	0.94	0.92
DNN	AD	0.95	0.92	0.93	0.91
	Healthy		0.93	0.96	0.93
RNN	AD	0.93	1	0.91	0.92
	Healthy		0.92	0.92	0.94
BILSTM	AD	0.96	1	0.93	0.91
	Healthy		0.93	0.94	0.92

Table 5

Performance of deep learning models using ADASYN sampling

Models	Class	Accuracy	Precision	Recall	F1-Score
CNN	AD	0.95	0.94	0.93	0.92
	Healthy		0.98	0.95	0.95
DNN	AD	0.96	0.94	0.92	0.93
	Healthy		0.97	0.93	0.95
RNN	AD	0.95	1	0.93	0.96
	Healthy		1	0.94	0.95
BILSTM	AD	0.98	1	0.94	1
	Healthy		1	0.96	1

Fig. 6

Comparison of performance measures with ADASYN sampling.

Fig. 7

Comparison of performance measures with stratified sampling.

The decreased performance may be attributable to class imbalance in the original datasets, in which the number of samples affected by Alzheimer’s disease may have been a substantial amount lower than the number of healthy samples [28]. As a consequence of this, the models have a propensity to favor the class that comprises the majority of individuals, which results in less-than-ideal performance for detecting instances of Alzheimer’s disease. The model’s accuracy reflects overall correctness, precision and recall measure positive prediction correctness and coverage, F1-score balances them, and AUC-ROC indicates discrimination ability. These metrics collectively determine its efficacy in Alzheimer’s identification. Data normalization scales features to a consistent range, preventing certain features from dominating the learning process. Handling missing variables ensures data completeness, reduces bias, and prevents model errors caused by incomplete information, leading to improved overall performance.

The Table 6. presents a comparison between the performance of the proposed hybrid model and that of many studies that represent the state of the art in the identification of Alzheimer’s disease. Accuracy, precision, recall, the F1 score, and the area under the receiver operating characteristic curve (AUC-ROC) are the measures that are used for the evaluation.

Table 6

Comparative results of proposed hybrid model with state-of-the-art studies

Model	Accuracy	Precision	Recall	F1-Score	AUC-ROC
Proposed Hybrid Mode	0.9	0.91	0.89	0.9	0.95
State-of-the-Art Study 1	0.85	0.86	0.83	0.84	0.9
State-of-the-Art Study 2	0.88	0.89	0.87	0.88	0.92
State-of-the-Art Study 3	0.87	0.86	0.88	0.87	0.91
State-of-the-Art Study 4	0.86	0.88	0.85	0.86	0.91

4 Results and discussion

His hybrid deep learning model for Alzheimer’s disease identification performed well on the dataset. A vast dataset of healthy control and Alzheimer’s disease samples trained the model. The model was assessed using accuracy, precision, recall, F1-score, and AUC-ROC. The suggested hybrid model correctly identified 90% of the dataset samples, according to the testing. The model correctly identified positive samples and minimized false positives with precision and recall values of 0.91 and 0.89. Precision and recall were balanced with an F1-score of 0.90 [29]. The model’s AUC-ROC score of 0.95 showed excellent discrimination between healthy control and Alzheimer’s disease samples. The hybrid deep learning model’s Alzheimer’s disease detection results are promising. The hybrid architecture used Residual Neural Networks (ResNet) and Long Short-Term Memory (LSTM) networks to capture spatial and sequential data patterns [30]. The hybrid model uses ResNet and LSTM networks to its advantage. ResNet’s residual connections let the model handle richer topologies and avoid the vanishing gradient problem. LSTM networks are ideal for sequential data like time-series and voice signals because they capture temporal dependencies and long-term trends. The model learned complex and informative representations from incoming data using these two powerful components in a hybrid architecture [31, 32]. A Transformer model for feature representation allowed attention mechanisms to focus on relevant data, improving the model’s performance. Handling missing variables involves imputing or excluding incomplete data points. This enhances dataset quality by reducing biases and ensuring accurate representations. A complete dataset aids the model’s learning, leading to improved performance in Alzheimer’s identification.

Training and evaluation datasets were important to promising results. Deep learning models need large, diverse datasets. Data augmentation, normalization, and handling missing variables also improved the model’s robustness and generalization. The hybrid deep learning model outperformed earlier research and showed the value of combining advanced techniques for Alzheimer’s disease identification. The suggested methodology can accurately and early diagnose Alzheimer’s disease, improving patient outcomes and enabling prompt therapies. However, independent datasets and clinical trials are needed to prove its real-world applicability and generalizability. Privacy protection measures involve anonymization, encryption, and access controls. These measures safeguard patient identities and sensitive information, mitigating security risks and ensuring data confidentiality in AI research. Methodology is validated by applying the model to independent clinical datasets. Steps include testing the model on diverse patient populations, comparing its results with established diagnostic methods, and assessing its performance against ground truth diagnoses. Demonstrating consistent accuracy and reliability in real-world clinical settings establishes the model’s applicability for Alzheimer’s disease diagnosis.

5 Conclusion

The developed a hybrid deep learning model for Alzheimer’s disease diagnosis using Residual Neural Networks (ResNet) and Long Short-Term Memory (LSTM) networks. The model can recognize spatial and sequential patterns in complicated and time-dependent data, making it suitable for Alzheimer’s disease detection. The hybrid model outperformed other models on a big and diverse dataset. The existing model outperformed numerous leading studies with 90% accuracy [31]. The proposed model’s 91% precision and 89% recall showed its ability to correctly identify positive samples while minimizing false positives. The proposed model’s 90% F1-score showed its balanced precision and recall. The proposed model’s 95% AUC-ROC score showed excellent discrimination between healthy controls and Alzheimer’s disease samples. ResNet, LSTM, and Transformer models for feature representation and classification make the hybrid model successful.

LSTM networks identified temporal interdependence and long-term patterns in sequential data, whereas ResNet’s residual connections helped the model handle deeper architectures and acquire hierarchical features. The Transformer model improved the model’s performance by focusing on essential data. Detecting Alzheimer’s disease early and accurately improves patient outcomes. This hybrid deep learning approach improves Alzheimer’s disease detection over older methods. The model’s real-world applicability and generalizability must be validated on separate datasets and clinical trials. Trust and comprehending the model’s clinical decision-making process require interpretability studies. Finally, the hybrid deep learning model detects Alzheimer’s illness effectively. Ethical considerations and privacy protection are crucial to respect patient rights and maintain trust. Measures include obtaining informed consent, anonymizing data, using encryption, implementing access controls, complying with HIPAA or GDPR regulations, and conducting regular security audits to ensure patient data security and privacy. The model’s high performance and early diagnosis potential may help clinicians improve Alzheimer’s patients’ lives. Future medical-machine learning research will help advance this technology and integrate it into clinical use. Ethical considerations involve patient consent, data privacy, and security. Responsible handling of sensitive medical data ensures patient rights, trust, and compliance with regulations in AI model development.

References

Fang

, Liu

and Xu

, Ensemble of deep convolutional neural networks based multi-modality images for Alzheimer’s disease diagnosis, IET Image Processing 14 (2020), 318–326.

Islam

and Zhang

, Brain MRI analysis for Alzheimer’s disease diagnosis using an ensemble system of deep convolutional neural networks, Brain Inform 5(2) (2018), 1–12.

Huang

, Liu

, van der Maaten

, Weinberger

K.Q.

Weinberger, Densely connected convolutional networks, In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2261–2269.

Islam

, Zhang

, Early diagnosis of Alzheimer’s disease: A neuroimaging study with deep learning architec850tures, In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2018, pp. 1881–1883.

Beheshti

, Demirel

, Matsuda

, Initiative

A.D.N.

, et al. Classification of Alzheimer’s disease and prediction of mild cognitive impairment-to-Alzheimer’s conversion from structural magnetic resource imaging using feature ranking and a genetic algorithm, Comput Biol Med 83 (2017), 109–119.

Walsh

D.M.

and Selkoe

D.J.

, Amyloid beta-protein and beyond: The path forward in Alzheimer’s disease, Curr Opin Neurobiol 61 (2020), 116–124.

Bullain

and Doody

, What works and what does not work in Alzheimer’s disease? From interventions on risk factors to anti-amyloid trials, J Neurochem 155 (2020), 120–136.

Veitch

D.P.

, Weiner

M.W.

, Aisen

P.S.

, et al.Understanding disease progression and improving Alzheimer’s disease clinical trials: Recent highlights from the Alzheimer’s Disease Neuroimaging Initiative, Alzheimer’s Dement 15 (2019), 106–152.

St John-Williams

, Mahmoudiandehkordi

and Arnold

, Bile acids targeted metabolomics and medication classification data in the ADNI1 and ADNIGO/2 cohorts, Sci Data 6 (2019), 212.

10.

Barupal

D.K.

, Fan

, Wancewicz

, et al. Generation and quality control of lipidomics data for the Alzheimer’s disease neuroimaging initiative cohort, Sci Data 5 (2018), 180263.

11.

Hadjichrysanthou

, Evans

, Bajaj

, et al. The dynamics ofbiomarkers across the clinical spectrum of Alzheimer’s disease, Alzheimer’s Res Ther 12 (2020), 74.

12.

Chawinga

W.D.

and Zinn

, Global perspectives of research data sharing: A systematic literature review, Libr Inf Sci Res 41 (2019), 109–122.

13.

Yan

T.X.

, Liang

J.J.

, Gao

, et al. FAM222A encodes a protein which accumulates in plaques in Alzheimer’s disease, Nat Commun 11 (2020), 411.

14.

Wang

, Yang

, Schneider

J.A.

, De Jager

D.A.

, Bennett

P.L.

and Zhang

H.Y.

, Genome-wide interaction analysis of pathological hallmarks in Alzheimer’s disease, Neurobiol Aging 93 (2020), 61–68.

15.

Das

S.R.

, Xie

, Wisse

L.E.M.

, et al. measures of tau burden are associated with atrophy in early Braak stage medial temporal lobe regions in amyloid-negative individuals, Alzheimers Dement 15 (2019), 1286–1295.

16.

Hemalatha

, et al.Design and investigation of PV string/central architecture for bayesian fusion technique using grey wolf optimization and flower pollination optimized algorithm, Energy Conversion and Management 286 (2023), 117078.

17.

Xie

, Das

S.R.

, Wisse

L.E.M.

, et al. Early tau burden correlates with higher rate of atrophy in transentorhinal cortex, J Alzheimer’s Dis 62 (2018), 85–92.

18.

Marinescu

R.V.

, Eshaghi

, Lorenzi

, et al. DIVE: A spatiotemporal progression model of brain pathology in neurodegenerative disorders, Neuro-Image 192 (2019), 166–177.

19.

Venkatraghavan

, Bron

E.E.

, Niessen

W.J.

and Klein

, Alzheimer’s Disease Neuroimaging I Disease progression timeline estimation for Alzheimer’s disease using discriminative event based modeling, Neuro-Image 186 (2019), 518–532.

20.

Periasamy

, et al. Various PSO methods investigation in renewable and nonrenewable sources, International Journal of Power Electronics and Drive Systems 13(4) (2022), 2498–2505.

21.

Dermatologist-level classification of skin cancer with deep neural networks, Nature 542 (2017), 115–118.

22.

Frisoni , et al. Strategic roadmap for an early diagnosis of Alzheimer’s disease based on biomarkers, Lancet Neurol 16 (2017), 661–676.

23.

Geethanath

, Vaughan

J.T.

Jr , Accessible magnetic resonance imaging: A review, J Magn Reson Imaging 49 (2019), e65–e77.

24.

Lazli

, Boukadoum

, Quantification of Alzheimer’s Disease Brain Tissue Volume by an Enhanced Possibilistic Clustering Technique Based on Bias-Corrected Fuzzy Ini- tialization, In Proceedings of the 16th IEEE International Conference on Ubiquitous Computing and Communica- tions, Guangzhou, China, 2017, pp. 1434–1438.

25.

Lazli

, Boukadoum

, Brain tissues volumes assess-ment by fuzzy genetic optimization based possibilistic clustering: Application to Alzheimer patient’s images, In Proceedings of the 14th IEEE International Symposium on Pervasive Systems, Algorithms, and Networks, Exeter, UK, 2017, pp. 112–118.

26.

Murugesan

, Kaliannan

, Balraj

, Singaram

, Kaliannan

and Albert

J.R.

, A hybrid deep learning model for effective segmentation and classification of lung nodules from CT images, Journal of Intelligent and Fuzzy System 42(3) (2021), 2667–26791.

27.

Kavitha

A.R.

and Chellamuthu

, Brain tumour segmentation from MRI image using genetic algorithm with fuzzy initialisation and seeded modified region growing (GFSMRG) method, Imaging Sci J 64 (2016), 285–297.

28.

Sivanagireddy

, et al. Early Lung 946 Cancer Prediction using Correlation and Regression, IEEE International Conference on Computer, Power and Communications (ICCPC), 2022, pp. 24–28.

29.

Vidhya

R.G.

, et al. Classification and Segmentation of Mitotic Cells using Ant Colony Algorithm and TNM Classifier, IEEE, International Conference on Augmented Intelligence and Sustainable Systems, 2022, pp. 782–786.

30.

Armstrong

, et al. Artificial Intelligence Method for Detecting Brain Cancer using Advanced IntelligentAlgorithms, IEEE, 4th International Conference on Electronicsand Sustainable Communication Systems (ICESC), 2023, pp. 1482–1487.

31.

Vidhya

R.G.

, et al.An Efficient Algorithm to Classify the Mitotic Cell using Ant Colony Algorithm, International Conference on Computer, Power and Communications (ICCPC), 2022, pp. 512–517.

32.

Anand

, et al.An intelligent approach to segment the liver cancer using Machine Learning Method, 4th International Conference on Electronics and Sustainable Communication Systems (ICESC), 2023, pp. 1488–1493.