Leveraging AI Models for Detection and Classification of Alzheimer’s Disease Using MRI and Clinical Data

1. Introduction

Alzheimer’s disease (AD), a progressive neurodegenerative disorder, presents a significant challenge for early diagnosis and effective management due to its complex and multifactorial nature. AD is the most prevalent type of dementia, affecting patients and their loved ones with a wide range of symptoms that significantly impair memory, reasoning, and social skills [1]. Before spreading to other outer layers of the brain, the illness first affects the hippocampus, which is the area of the brain in charge of memory and learning [2]. Early on, people may have trouble recalling recent events, such appointments or discussions. Conversely, it gets harder to remember important details such as one's name and relatives [3].

Jack et al. [4] shed light on the fundamental mechanisms of AD, identifying its key pathological characteristics as amyloid deposits, tau protein abnormalities, and neurodegeneration. These three core pathological features play a crucial role in prediction, diagnosis, and treatment of AD. Prior to the extensive use of artificial intelligence (AI) in healthcare, traditional methods for testing AD relied on a variety of techniques. Tools like the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MOCA) were employed to evaluate and score a patient's cognitive function, helping to assess their cognitive performance levels [5]. With advancements in technology, methods such as magnetic resonance imaging (MRI), positron emission tomography (PET), diffusion tensor imaging (DTI), biomarkers, and cerebrospinal fluid (CSF) analysis are increasingly utilized for detecting AD, as they eliminate the influence of subjective factors [6]. MRI technology uses a strong magnetic field and harmless radio waves to generate high-resolution brain images, aiding physicians in observing the brain structure and detecting potential abnormalities [7]. MRI is crucial in diagnosing Alzheimer’s disease as it provides high-resolution, non-invasive imaging of brain structures, enabling the detection of early signs of neurodegeneration, such as hippocampal atrophy and cortical thinning, which are key indicators of the disease’s progression [8]. In the early stages of Alzheimer’s disease, the pathological features are less pronounced, making brain imaging methods like MRI potentially insufficiently sensitive for accurate prediction of the condition [9].

AI can enhance the sensitivity of brain imaging techniques, such as MRI, by leveraging advanced algorithms to detect subtle patterns and early-stage biomarkers of Alzheimer’s disease that might otherwise go unnoticed through traditional analysis, thereby improving early diagnosis and intervention strategies [10]. Tackling the challenges of diagnosing and treating complex conditions such as AD has driven a growing interest in leveraging advanced technologies to improve clinical outcomes. AI, particularly through machine learning (ML) and deep learning (DL), holds tremendous promise in revolutionizing AD diagnostics and care. By analyzing vast amounts of medical data, AI systems can detect subtle patterns and early biomarkers that traditional methods might miss, enabling earlier diagnosis and more personalized intervention strategies. The concept of AI was first introduced by John McCarthy in 1956, who defined it as the use of computer systems to replicate human intelligence and critical reasoning [11].

In healthcare, AI is categorized into two main domains: virtual and physical. The virtual domain encompasses ML and DL [12]. Machine learning refers to a system’s ability to autonomously learn from data without explicit programming [11]. It includes four primary methodologies: supervised learning, unsupervised learning, reinforcement learning, and active learning [13]. Supervised learning involves analyzing labeled input data to uncover patterns, utilizing models such as Bayesian inference, decision trees, linear discriminants, support vector machines, logistic regression, and artificial neural networks [14]. Deep learning, a more advanced subset of ML, employs multiple interconnected layers to extract features and optimize model performance [15].

AI technologies aim to develop systems and robots capable of performing tasks like pattern recognition, decision-making, and adaptive problem-solving—capabilities traditionally associated with human intelligence [16]. Advances in computational power, combined with innovations in machine learning techniques and neural networks, have accelerated progress in AI [17]. As a subset of AI, ML focuses on training computers to analyze large datasets, identify trends, and apply these insights for predictions or decisions [16]. AI has demonstrated transformative potential across fields such as natural language processing, autonomous vehicles, healthcare, and image recognition. In AD research, it excels at rapidly analyzing complex datasets, identifying patterns imperceptible to humans, and providing highly accurate predictions, thereby advancing the understanding and management of the disease [18], [19]. DL is centered around advanced neural network architectures, including Convolutional Neural Networks (CNNs) [20] and Artificial Neural Networks (ANNs) [21].

CNNs are a specialized type of ANN designed to process and analyze visual data, such as images. Unlike ANNs, CNNs leverage convolutional layers that apply filters (kernels) to extract spatial and hierarchical features like edges, textures, and shapes [22]. These layers are followed by pooling layers, which reduce the spatial dimensions and improve computational efficiency [23]. Fully connected layers at the end of the network use the extracted features to make predictions [24]. CNNs excel at tasks like image recognition, object detection, and medical imaging due to their ability to capture spatial relationships and patterns in data [25]. ANNs are inspired by the structure and function of the human brain, consisting of layers of interconnected nodes (neurons) [26]. These nodes process input data by applying weights, biases, and activation functions, which enable the network to learn and make predictions. ANNs typically have an input layer (to receive data), one or more hidden layers (where computations and feature extraction occur), and an output layer (to generate predictions) [27]. They are versatile and suitable for a wide range of tasks, such as classification, regression, and time-series analysis.

This study explores the application of AI in enhancing the prediction and diagnosis of AD by leveraging two distinct AI models: ANNs and CNNs. By utilizing ANNs, which are well-suited for handling structured data, and CNNs, which excel at processing and analyzing visual data like medical imaging, this research aims to address the challenges of early detection and improve diagnostic accuracy. Both models were trained and evaluated using relevant datasets to identify patterns, biomarkers, and structural abnormalities associated with AD. The integration of these complementary AI approaches demonstrates their potential in providing robust, reliable tools for predicting and diagnosing AD, ultimately contributing to more effective clinical interventions and patient outcomes.

Statement of Significance
Problem or Issue:	Early and accurate diagnosis of Alzheimer’s disease (AD) remains a significant challenge due to the complex nature of the disease and the limitations of traditional diagnostic methods in detecting subtle patterns in clinical and imaging data.
What is Already Known:	Machine learning models, including artificial neural networks (ANNs) and convolutional neural networks (CNNs), have been applied individually to clinical and imaging datasets for AD diagnosis. However, these approaches often lack integration and struggle to combine multimodal data effectively.
What This Paper Adds:	This paper introduces a dual-model framework that integrates an ANN for clinical data analysis and a CNN for imaging data classification, addressing the need for a comprehensive diagnostic approach. The ANN achieved 87.08% accuracy in risk assessment, while the CNN excelled with 97% accuracy in classifying AD stages. Grad-CAM visualizations further enhance interpretability, highlighting regions critical to diagnosis. This study demonstrates the potential of combining clinical and imaging data to improve early detection and classification of AD, paving the way for scalable, AI-driven diagnostic systems.

2. Datasets and Symptom Analysis

In this study, we developed two distinct AI models to predict Alzheimer's disease stages and assess their potential severity. The first model employed a CNN trained on 4,876 MRI brain images, while the second model utilized an ANN trained on clinical data collected from 1,200 patients. The primary aim was to create an integrated diagnostic framework capable of accurately classifying Alzheimer's disease into four categories: mild demented, moderate demented, non-demented, and very mild demented.

The MRI dataset consisted of 4,876 labeled brain images distributed across the four categories. To ensure uniformity and compatibility with the CNN model, the images were preprocessed by resizing them to a standard resolution of 256x256 pixels and normalizing pixel intensity values to enhance computational efficiency while preserving critical diagnostic features. The CNN model predicts the category of the uploaded MRI image along with a probability score for each class, providing insights into the likelihood of a specific diagnosis. Additionally, Grad-CAM was employed to generate heatmap overlays on input images, highlighting regions critical to the model’s predictions. These heatmaps enhance visual interpretability, making the diagnostic system more transparent and clinically relevant.

The clinical dataset comprised data from 1,200 patients, capturing 31 features related to demographics, lifestyle, genetic predispositions, and symptomatic factors associated with Alzheimer’s disease. These features included:

• Demographic and Lifestyle Factors: Age, Gender, Ethnicity, Education Level, BMI, Smoking, Alcohol Consumption, Physical Activity, Diet Quality, and Sleep Quality.
• Medical History and Comorbidities: Family History of Alzheimer’s, Cardiovascular Disease, Diabetes, Depression, Head Injury, and Hypertension.
• Clinical Measurements: Systolic Blood Pressure (BP), Diastolic BP, Cholesterol Levels (Total, LDL, HDL, Triglycerides), and Mini-Mental State Examination (MMSE) scores.
• Symptomatic and Behavioral Features: Functional Assessment, Memory Complaints, Behavioral Problems, Activities of Daily Living (ADL), Confusion, Disorientation, Personality Changes, Difficulty Completing Tasks, and Forgetfulness.

To ensure compatibility with the ANN model, the dataset underwent preprocessing, including standardization of feature values to improve model training consistency. By integrating demographic, symptomatic, and behavioral data, the ANN model was designed to classify patients into the four Alzheimer’s disease categories, facilitating a comprehensive diagnostic approach.

3. Machine Learning Model

A CNN was developed using Python and the TensorFlow library to classify MRI images into four categories. Preprocessing steps ensured uniformity, with all images resized to 256x256 pixels and normalized pixel intensity values. The CNN architecture consisted of five convolutional layers with The Rectified Linear Unit (ReLU) activation functions, containing 64, 128, 128, 64, and 64 filters, respectively, followed by max-pooling layers to reduce dimensionality while retaining significant features. A Flatten layer was included to prepare the data for fully connected layers. The dense layer consisted of 64 neurons with a ReLU activation function, followed by a final dense layer with 4 neurons and a SoftMax activation function to predict the probabilities for the four categories. Dropout layers were incorporated to mitigate overfitting, and batch normalization was employed to stabilize and accelerate training.

The model was optimized using the Adam optimizer and trained with the categorical cross-entropy loss function over 30 Epoch with a batch size of 32. Data augmentation techniques, such as random rotations, flips, zooms, and shifts, were applied to enhance robustness and simulate real-world imaging conditions. Performance metrics, including accuracy, precision, recall, and F1 score, alongside confusion matrices, were used to evaluate the model’s effectiveness.

The ANN model employed feed-forward architecture with input, hidden, and output layers. The input layer processed 31 clinical features, followed by a dense hidden layer with 64 neurons using a ReLU activation function. The output layer comprised 2 neurons with a sigmoid activation function, designed for binary classification tasks. The ANN was trained using the Adam optimizer and binary cross-entropy loss function. Standard evaluation metrics, including accuracy, precision, recall, and F1 score, were utilized to evaluate the model’s performance. Confusion matrices provided detailed insights into the distribution of predictions relative to the true labels, offering a clear understanding of the model’s classification accuracy.

The dual-model approach, combining the CNN for imaging data and the ANN for clinical data, offers a robust diagnostic framework. CNN excels in image-based analysis, while the ANN integrates comprehensive patient-specific data. Together, these models highlight the potential of deep learning techniques to enhance diagnostic accuracy and provide valuable insights into Alzheimer’s disease classification and management.

4. Experimental Results

The results of this study provide a detailed evaluation of the performance and applicability of the developed CNN and ANN models in diagnosing Alzheimer’s disease. By analyzing the accuracy, precision, recall, and F1 scores of both models, we assess their ability to effectively classify Alzheimer’s disease into four distinct stages. Additionally, confusion matrices and visual explanations generated by Grad-CAM enhance the interpretability and transparency of the CNN model’s predictions. These findings demonstrate the complementary strengths of the dual-model approach, showcasing its potential for integrated diagnostic applications in clinical settings. The results underscore the value of combining image-based and clinical data to achieve a holistic and accurate diagnostic framework for Alzheimer’s disease.

Figure 1 illustrates the performance of a CNN trained to detect Alzheimer's disease, displaying metrics over 30 epochs. The left plot shows the training and validation accuracy. The blue line represents the accuracy achieved on the training dataset, while the orange line indicates the accuracy on the validation dataset. Both lines steadily increase and converge, demonstrating that the model's predictions improve over time. The consistent alignment between the training and validation accuracy suggests that the model generalizes well and performs effectively on unseen data.

The right plot displays the training and validation loss, with the blue line representing the loss on the training dataset and the orange line indicating the loss on the validation dataset. Both loss values decrease over the epochs, reflecting successful learning by the model. The convergence of the loss values, along with minimal signs of overfitting, highlights the model's robustness and reliability in detecting Alzheimer's disease. Together, these plots confirm that the CNN model has been trained effectively, achieving high accuracy and low loss while maintaining good generalization to new data.

Figure 2 represents the Receiver Operating Characteristic (ROC) curve for a multi-class classification problem in the context of Alzheimer's disease detection using a Convolutional Neural Network (CNN) model. The ROC curve plots the True Positive Rate (Sensitivity) against the False Positive Rate (1 - Specificity) for each class, providing a visualization of the model's performance for distinguishing between the four classes of Alzheimer's disease. Each curve demonstrates how the sensitivity and specificity trade-off changes at different classification thresholds. The closer the curve is to the top-left corner of the plot, the better the model's performance. The overlapping or closely aligned curves suggest high classification accuracy across all classes, as reflected by the minimal gaps between the curves. This figure highlights the model's robustness and effectiveness in distinguishing between various stages of Alzheimer's disease.

Table 1 showcases the performance metrics of a Convolutional Neural Network (CNN) model developed to classify the progression of Alzheimer's disease into four categories. These metrics, including accuracy, precision, recall, F1-score, and Matthews Correlation Coefficient (MCC), provide a comprehensive overview of the model's effectiveness. The accuracy of the model is 0.97, indicating that 97% of all predictions were correct. This high accuracy reflects the model's overall reliability across all classes. The weighted precision, at 0.98, measures the proportion of true positive predictions out of all positive predictions, showing the model’s reliability in identifying true instances for each class. Similarly, the weighted recall is 0.97, which evaluates the proportion of true positive predictions among all actual positive instances, highlighting the model's sensitivity. The weighted F1-score, a balance between precision and recall, is 0.98, suggesting that the model achieves an excellent trade-off between false positives and false negatives. Furthermore, the Matthews Correlation Coefficient (MCC), which assesses the overall quality of classification, stands at 0.96, reflecting strong agreement between predicted and true labels.

For the "Mild Demented" class, the model achieves a precision of 0.99, recall of 0.97, and F1-score of 0.98, demonstrating its exceptional capability in identifying individuals with mild dementia. For the "Moderate Demented" class, the precision is slightly lower at 0.81, but the recall reaches 1.00, yielding an F1-score of 0.90. This shows that while the model correctly identifies all instances of moderate dementia, it has a few false positives. For the "Non-Demented" class, the model performs nearly perfectly, with precision and recall both at 0.99, resulting in an F1-score of 0.99. The "Very Mild Demented" class also shows strong performance, with precision at 0.99, recall at 0.95, and F1-score at 0.97, indicating high reliability. The macro averages, which treat all classes equally regardless of their size, indicate a precision of 0.95, recall of 0.98, and F1-score of 0.96. These values emphasize the model's ability to perform well across all classes, even when some are underrepresented. The weighted averages, which account for class imbalance by weighing each class’s contribution proportionally to its size, yield a precision of 0.98, recall of 0.97, and F1-score of 0.98. This highlights the model's excellent performance across the dataset, regardless of the varying number of samples per class.

Figure 3 demonstrates the confusion matrix for the CNN model used to classify Alzheimer's disease stages. The confusion matrix provides a detailed view of the model's predictions compared to the actual labels, highlighting both correct and incorrect classifications. Each row corresponds to the true class labels, while each column represents the predicted class labels. For the "Mild Demented" class, the model correctly classifies 143 out of 148 samples, with only 5 samples being misclassified as "Moderate Demented." Notably, none of the "Mild Demented" samples were misclassified as "Non-Demented" or "Very Mild Demented." The "Moderate Demented" class demonstrates perfect performance, as all 39 samples are correctly classified, with no misclassifications observed. Similarly, for the "Non-Demented" class, the model achieves near-perfect results, correctly classifying 173 out of 174 samples, with only one sample misclassified as "Very Mild Demented." The "Very Mild Demented" class also shows strong performance, with 144 out of 151 samples correctly classified. However, there are a few misclassifications in this class, with 4 samples labeled as "Moderate Demented" and 2 as "Non-Demented.".

Figure 4 provides representative examples of the CNN model's predictions for Alzheimer's disease classification based on MRI images. Each sub-image includes the actual label, predicted label, and the confidence score of the prediction, showcasing the model's ability to classify different stages of Alzheimer's disease with high accuracy.

On the left side, the first two rows show "Non-Demented" cases, where both the actual and predicted labels are "Non-Demented." The confidence score for these predictions is 100%, reflecting the model's absolute certainty. These images indicate the structural patterns that the model associates with the absence of dementia. Moving to the middle section, the images depict cases labeled as "Mild Demented," where the model correctly predicts the same class with a confidence of 100%. These samples demonstrate the model's ability to identify the subtle features of mild dementia from the MRI scans. On the right side, the figure presents cases labeled as "Very Mild Demented." Again, the model correctly predicts the same class with confidence scores either at 100% or very close (e.g., 99.99%). These predictions highlight the model's precision in distinguishing between different early stages of dementia.

Figure 5 illustrates a Grad-CAM (Gradient-weighted Class Activation Mapping) visualization for the CNN model's prediction of an MRI image classified as "Very Mild Demented." Grad-CAM highlights the regions of the brain scan that contributed most significantly to the model's decision, providing an interpretable explanation of the classification process. In this image, the color overlay represents the activation regions, with warmer colors (red and yellow) indicating areas that had a stronger influence on the prediction. Cooler colors (green and blue) represent less relevant regions. The highlighted regions correspond to structural features that the model associates with the "Very Mild Demented" stage, emphasizing the key parts of the brain that distinguish this condition.

Figure 6 shows the ANN model's accuracy during the training and validation processes over 18 epochs. The blue line represents the accuracy achieved on the training dataset, while the orange line reflects the accuracy on the validation dataset. Initially, the accuracy for both the training and validation datasets increases rapidly, indicating that the model is learning to distinguish features effectively. By around the 5th epoch, the validation accuracy starts to stabilize, reaching a plateau at approximately 85%. The training accuracy, on the other hand, continues to improve and eventually surpasses 95%. The gap between the training and validation accuracy after the 5th epoch indicates a slight overfitting, where the model performs better on the training data than on unseen validation data. However, the consistently high validation accuracy suggests that the model generalizes well to new data, making it a reliable tool for Alzheimer's disease classification.

Confusion matrix summarizes ANN model performance of a binary classification model designed to detect Alzheimer's disease in Figure 8. The matrix outlines the relationship between the true and predicted labels. The rows correspond to the actual labels, where "0" represents cases without Alzheimer's and "1" represents cases with Alzheimer's. The columns represent the predicted labels, with "0" indicating predictions of "No Alzheimer's" and "1" indicating predictions of "Alzheimer's.". The top-left cell shows that the model correctly identified 144 cases as "No Alzheimer's," demonstrating its ability to accurately classify these instances (true negatives). Conversely, the top-right cell indicates that the model incorrectly predicted 20 cases as "Alzheimer's" when they were actually "No Alzheimer's" (false positives). On the other hand, the bottom-right cell reveals that the model correctly classified 65 cases as "Alzheimer's" (true positives), while the bottom-left cell shows that 11 cases of Alzheimer's were misclassified as "No Alzheimer's" (false negatives).

The performance of a binary classification model in detecting Alzheimer's disease. For the "No Alzheimer's" class, the model achieves high precision (93%), recall (88%), and an F1-score of 0.90, reflecting strong performance. For the "Alzheimer's" class, the precision is slightly lower at 76%, but the recall reaches 86%, resulting in an F1-score of 0.81. The weighted averages for precision, recall, and F1-score are 0.88, 0.87, and 0.87, respectively, showing a balanced performance across both classes. With an overall accuracy of 87.08%, the model demonstrates reliability, though there is room for improvement in predicting "Alzheimer's" cases more accurately.

5. Discussion

In this study, we developed and evaluated two distinct artificial intelligence models, an ANN and a CNN, for predicting Alzheimer's disease stages and assessing its severity. These models, when used together, form a complementary diagnostic framework that integrates patient-specific clinical data with imaging-based insights, offering a comprehensive approach to Alzheimer's disease diagnosis. Similar hybrid approaches have been proposed in previous research, demonstrating the effectiveness of combining clinical and imaging data to improve diagnostic precision for neurodegenerative diseases [28]. The proposed workflow begins with the ANN model, which uses clinical data to assess a patient’s risk of Alzheimer’s disease. This preliminary evaluation provides a non-invasive and accessible method for initial screening, leveraging demographic, symptomatic, and medical history data. Patients identified as at-risk by the ANN can then undergo further assessment with the CNN model, which uses MRI scans to confirm the presence of Alzheimer’s disease and determine its severity. The CNN also provides detailed classification into disease stages—mild demented, moderate demented, very mild demented, or non-demented—enhancing diagnostic precision and clinical relevance.

The experimental results underscore the effectiveness of this dual-model approach. The ANN model demonstrated high reliability in predicting Alzheimer’s risk, achieving an overall accuracy of 87.08%. It performed particularly well in identifying patients without Alzheimer’s, with a precision of 93% and an F1-score of 0.90. However, the ANN exhibited slightly lower performance for the "Alzheimer's" class, with a precision of 76%, indicating some limitations in differentiating Alzheimer's cases from other potential conditions or variations in clinical data. These results align with findings from previous studies that emphasize the challenges of using clinical data alone to diagnose Alzheimer's disease due to overlapping symptoms with other conditions [29]. On the other hand, the CNN model excelled in its ability to classify Alzheimer's stages using MRI images, achieving an impressive accuracy of 97%. The use of CNNs for neurodegenerative disease classification has been widely validated in the literature, with similar studies achieving high accuracy through optimized architecture and data augmentation techniques [30]. The model demonstrated nearly perfect performance in distinguishing non-demented cases and identifying mild dementia, with precision and recall scores exceeding 95% for these categories. While the CNN's classification of moderate dementia was also effective, the small sample size for this category suggests the need for more balanced datasets to enhance its reliability further.

The integration of ANN and CNN models offers several advantages. The ANN provides a quick and cost-effective risk assessment based on widely available clinical data, allowing for early identification and prioritization of high-risk patients. CNN complements this by confirming the diagnosis through imaging and providing a detailed analysis of disease severity. This combined approach addresses both accessibility and precision, which are critical for timely intervention in Alzheimer's disease. Previous research has highlighted those multimodal diagnostic approaches, which integrate multiple data types, significantly improve diagnostic accuracy compared to single-modality systems [31]. Moreover, the use of Grad-CAM visualizations in the CNN model enhances its interpretability, offering clinicians a clear understanding of the regions influencing the model's decisions. This transparency is particularly valuable in medical applications, where trust in AI-driven outcomes is essential [32].

Despite these strengths, there are limitations to consider. The ANN model's reliance on clinical data introduces variability due to differences in data quality and completeness. This limitation is commonly reported in studies using electronic health records or self-reported data, which can be prone to errors and inconsistencies [33]. The CNN, while highly accurate, requires access to MRI imaging, which may not be readily available in all clinical settings. Additionally, both models showed some degree of overfitting, particularly in cases where the training datasets were imbalanced, indicating a need for more robust training techniques or larger, more diverse datasets. The dual-model approach also introduces a dependency on multiple diagnostic stages, which, while comprehensive, may increase the time and resources required for complete diagnosis.

Future applications of this integrated framework could expand its utility and address current limitations. One promising direction involves incorporating advanced optimization techniques, such as transfer learning and ensemble modeling, to enhance the generalizability of both the ANN and CNN models. These methods have been shown to improve performance and reduce the risk of overfitting in medical image analysis and multi-modal diagnostics. Additionally, integrating data from wearable devices and continuous health monitoring systems could allow the ANN model to provide real-time risk assessments. Recent studies have demonstrated the potential of wearable technology in capturing early biomarkers of neurodegenerative diseases, which could significantly aid in the early detection of Alzheimer's [34]. Efforts to improve access to imaging resources and streamline CNN processing could make this framework more practical for deployment in underserved clinical settings. The development of lightweight CNN models or cloud-based diagnostic platforms could further enhance scalability and accessibility, as evidenced by similar initiatives in other healthcare domains. Ultimately, by bridging clinical and imaging data, this dual-model system offers an innovative and scalable solution for Alzheimer’s disease diagnosis, paving the way for improved patient care and early intervention strategies.

6. Conclusion

This study proposes a dual-model framework integrating an ANN and a CNN model to enhance the diagnosis and classification of Alzheimer’s disease. The ANN provides a quick and accessible method for assessing a patient’s risk using clinical data, while the CNN confirms the diagnosis and determines disease severity through MRI imaging. Together, these models offer a comprehensive diagnostic solution, combining clinical and imaging data for greater accuracy. The ANN achieved an accuracy of 87.08%, effectively identifying high-risk patients, and the CNN excelled with 97% accuracy in classifying Alzheimer’s stages. Grad-CAM visualizations further enhanced CNN’s interpretability, making the system clinically relevant. While the framework demonstrated robust performance, limitations such as variability in clinical data and reliance on MRI imaging suggest areas for improvement.

Future advancements, such as integrating wearable technology for continuous monitoring and optimizing the CNN for scalability, could expand the framework’s utility. This dual-model system underscores the potential of AI in transforming Alzheimer’s diagnostics, paving the way for early interventions and improved patient care.