Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 799 https://internationalpubls.com Deep Learning-Driven Prognostic Modeling for Alzheimer’s Disease Identification Awais Ahmed1 Mr. Rizwan Uz Zaman2 Mrs. MD Asma3 1 Research Scholar, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana 2Assistant professor, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana 3Associate professor, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana Article History: Received: 12-01-2025 Revised: 15-02-2025 Accepted: 01-03-2025 Abstract Alzheimer’s disease is an genuine neurodegenerative sickness that impacts brain memory fundamentally in developed people. Alzheimer’s contamination happens around the world and fundamentally impacts people developed more prepared than 65 a long time. Early assurance for correct area is required for this ailment. Manual assurance by prosperity masters is botch slanted and time exhausting due to the broad number of patients showing with the illness. Diverse procedures have been associated to the assurance and classification of Alzheimer’s disease but there's a require for more precision in early conclusion courses of action. The appear proposed in this examine prescribes a significant learning-based course of action utilizing DenseNet-169 and ResNet-50 CNN plans for the assurance and classification of Alzheimer’s disease. The proposed illustrate classifies Alzheimer’s disease into Non-Dementia, Especially Tender Dementia, Tender Dementia, and Coordinate Dementia. The DenseNet-169 designing beated inside the planning and testing stages. The planning and testing precision values for DenseNet-169 are 0.977 and 0.8382, though the exactness values for ResNet-50 were 0.8870 and 0.8192. The proposed illustrate is usable for real-time examination and classification of Alzheimer’s disease Keywords: Alzheimer's disease; artificial intelligence; classification; deep learning; machine learning; magnetic resonance imaging; neuroimaging; positron emission tomography. 1. INTRODUCTION Alzheimer’s ailment (Promotion) might be a sort of miserable brain disease due to neurodegeneration. Notice is appear around the world. Promotion is characterized by β- amyloid (Aβ), which contains extracellular plaques and tau-containing intracellular neurofibrillary tangles Cognitive capacity[1] clutter is the major side impact due to Promotion. This sickness is more transcendent in developed people, commonly impacting those developed 65 or more prepared; 10% of cases are early onset happening in people more young than 65. Notice as well impacts lingo,[2] thought, comprehension, considering, and memory. Specialists can be careful of patients persevering from this disease’s side impacts. The rot in cognitive Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 800 https://internationalpubls.com capacity happens due to dementia which impacts day by day works out. Notice is the preeminent common sort of dementia, bookkeeping for nearly two-thirds of the cases due to age factors. In 2020, Promotion was the seventh driving[3] cause of passing inside the Joined together States of America. Promotion incorporates a few medicines to create strides the signs, but there's no genuine treatment to recover Notice sorts are classified as Non-Dementia, Very- Mild Dementia[3], Delicate Dementia, and Coordinate Dementia. Concurring to Notice examination, these stages are characterized concurring to other examine strategies and are particular from the DSM-5 classification of Promotion Signs of Notice depend upon the orchestrate of the illness.[4] The first common and especially to start with side impact is short- term memory hardship. As well, a tongue clutter is common in Promotion patients. Promotion side impacts routinely do not show up inside the early stages, which is the major boundary to genuine treatment.[5] As there's no true blue treatment for Notice, early conclusion engages potential treatment to cover the early stages. In any case, early assurance is challenging due to the presentation of minor side impacts and presently and after that it isn't conceivable to suitably recognize the signs[6]. Commonly, a neuropsychological examination is utilized for the early assurance of Notice. Clinicians are careful for suitably analyzing the Notice calm, but the manual examination is gloomy and takes time for the number of patients symptomatic for the ailment Helpful pros are careful for Notice conclusion, in a idealize world[7],[8] found inside the early stages. In any case, appallingly, due to the sweeping volume of data in therapeutic pictures, and the tremendous number of patients, it is boundless to accurately and quickly analyze them physically. Each clinician or restorative ace physically analyzes numerous of the records and gives an examination based on their experience and data. 2. LITERATURE REVIEW 1. Nanomaterials for the Treatment and Conclusion of Alzheimer’s Disease: An Chart Bilal M., Barani M., Sabir F., Rahdar A., Kyzas G. Z. (2020): This wide overview examines how advanced nanomaterials—such as appealing nanoparticles, carbon nanotubes, quantum bits, and polymeric nanocapsules—are being planned as sensitive expressive devices and potential accommodating pros for Alzheimer’s disease (Notice). The article looks at applications amplifying from nanocarrier engaged calm movement over the blood-brain obstacle to nanobiosensors able of distinguishing Notice biomarkers. It emphasizes that such nanotechnology stages have the potential to move forward early ailment revelation, progress centered on treatment transport, and address current gaps in Alzheimer’s diagnostics and therapeutics 2. Alzheimer’s Sickness Classification Utilizing Trade Learning Budhiraja I., Garg D. This think almost benchmarks the practicality of diverse significant learning architectures— VGG 19, Start V3, ResNet 50, DenseNet 169, and vanilla CNNs—on a multi-class MRI dataset for Promotion (crossing Non Dementia, Outstandingly Tender, Tender, Coordinate stages). Leveraging pre-trained ImageNet weights and fine-tuning for MRI data, the makers report Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 801 https://internationalpubls.com promising classification comes approximately. They as well look at challenges posed by compelled labeled therapeutic data and propose methods for trade learning to overcome these obstacles . 3. Examination of Brain Sub-Regions Utilizing Optimization Procedures and Significant Learning Technique in Alzheimer Ailment Chitradevi D., Prabha S. Centering on specific brain locale, this work utilizes picture division (utilizing thresholding and U-Net) and optimization techniques like innate calculations and particle swarm optimization to move forward Notice disclosure. Morphological highlights are removed and energized into machine/deep learning classifiers to recognize Promotion patients from controls. Comes almost show up progressed symptomatic precision and offer bits of information into utilitarian organize alterations, appearing a framework that planning optimization over preprocessing, classification, and treatment orchestrating . 4. Significant Learning-Driven Alzheimer’s Ailment Classification: Custom CNN and Pretrained Structures for Exact MRI Examination Santos D.F. et al. (2023) A novel classification pipeline is proposed, combining a custom-built CNN with pre-trained models (DenseNet121, VGG16, InceptionV3, ResNet50) to recognize Promotion stages through MRI imaging. Assessed on multiclass datasets, the cross breed demonstrate accomplished 98.18% exactness, beating person design baselines. . 5. Classification of Alzheimer’s Disease Utilizing CNN with Trade Learning and Weighted Incident Oktavian M.W., Yudistira N., Ridok A. (2022) Tending to issues of lesson lopsidedness and dataset hindrances, this think approximately leverages trade learning with ResNet 18 and a weighted incident work to classify Alzheimer’s contamination stages. 3. PROPOSED SYSTEM We propose a solid CNN-based framework utilizing DenseNet-169 and ResNet-50 for multi- stage Alzheimer’s contamination conclusion from MRI checks. The system begins with comprehensive picture preprocessing—applying clamor diminishment methodologies like CLAHE or anisotropic dispersal, and brain division by recommends of K-means or U-Net—to progress solidify clarity . PreprocessedImages are at that point classified utilizing two unmistakable basic learning structures fine-tuned on a four-class MRI dataset (Non Dementia, Surprisingly Smooth, Smooth, Encourage Dementia). DenseNet-169 stands out, fulfilling organizing and testing correctnesses of 97.7% and 83.8%, autonomously, compared to 88.7% and 81.9% for ResNet-50 The pipeline sets explainability layers (e.g., Grad-CAM heatmaps) Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 802 https://internationalpubls.com to highlight illustrative brain locale, moving forward diagram straightforwardness and clinician confidence . Coordinated for real-time clinical course of action, the system sets data broadening inside the center of orchestrating and keeps up a input circle for diagram redesigns. This arranging offers a clinically down to soil, interpretable contraption for rectify, multi-stage Alzheimer’s classification from MRI input. 4. SYSTEM ARCHITECTURE Fig 5.1 System Architecture The system retrieves 3D MRI volumes, applies operations such as slice-by-slice extraction, intensity normalization, registration, and segmentation to produce standardized 2D/3D inputs fig 5.1 for deep CNN training (e.g., DenseNet, ResNet architectures). Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 803 https://internationalpubls.com 5. RESULTS AND DISCUSSION Observations for the Given Dementia Classification Bar Chart: Class Imbalance: The dataset is highly imbalanced. The majority class is "NonDemented" with over 2500 instances, followed by "VeryMildDemented" (~1800), while "MildDemented" and "ModerateDemented" have significantly fewer instances, especially the latter. Risk of Biased Learning: The imbalance may lead to machine learning models being biased towards the more frequent classes, resulting in poor performance for minority classes like "ModerateDemented". Need for Resampling: Techniques such as oversampling, undersampling, or synthetic data generation (e.g., SMOTE) might be necessary to address the imbalance and ensure fair learning. Real-World Reflection: The distribution likely reflects real-world trends where early or non-demented conditions are more common than moderate dementia in screened populations. Model Evaluation Consideration: Accuracy alone would be misleading; metrics like F1-score, AUC, and confusion matrix analysis are crucial for proper evaluation, especially for the underrepresented classes. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 804 https://internationalpubls.com General Observations Decreasing Trend: Both training and validation losses decrease consistently over epochs, indicating that the model is learning and improving over time. Early Overfitting Risk: In the initial epochs (up to around epoch 10), the validation loss fluctuates more than the training loss, but the gap between them remains manageable. Noise in Validation Loss: The validation loss shows noticeable fluctuations, especially compared to the smoother training loss, suggesting some variance or instability in validation performance. Training vs. Validation Loss Gap: After epoch 20, the training loss continues to drop steadily, while the validation loss plateaus and shows irregular spikes, indicating potential overfitting. Model Performance Implications The training process is effective, as the training loss steadily reduces with epochs. The validation loss fluctuates, which could be due to: A small or imbalanced validation dataset. Noisy labels or inconsistent input quality. Model starting to overfit to training data after a certain point. Recommendations Early Stopping: Consider using early stopping around epoch 25–30 to prevent overfitting. Regularization Techniques: Apply dropout, L2 regularization, or data augmentation to improve generalization. Cross-validation: Use k-fold cross-validation to verify model stability across different splits. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 805 https://internationalpubls.com Validation Set Review: Check the quality and balance of the validation dataset to ensure consistent performance measurement. Contextual Insight: In the context of Alzheimer’s detection, model generalization is critical. Overfitting could lead to misclassification of patient brain scans, resulting in false positives/negatives. Ensuring stable validation loss and balanced model performance is vital for clinical reliability and trust in AI- based diagnostic tools. The graph shows training and validation accuracy over 50 epochs for Alzheimer’s disease detection. Training accuracy steadily improves, reaching over 95%, indicating effective learning. Validation accuracy also increases but fluctuates slightly, stabilizing above 90%, suggesting good generalization with some variance. The consistent gap between training and validation accuracy after epoch 30 hints at mild overfitting. Overall, the model performs well with high predictive accuracy, but incorporating regularization or early stopping may further enhance validation stability and real-world performance. This level of accuracy is promising for clinical applications where reliable early detection is critical. 7. CONCLUSIONS AND FUTURE WORK Conclusion: It has been chosen that Alzheimer’s defilement is an hopeless neurodegenerative affliction that impacts brain memory, especially interior the elderly. Owing to the colossal number of patients, it is stunning to perform manual affirmation capably and thriving stars make goofs within the middle of assessment due to time destinations and the bother of the methodology. Particular strategies are utilized to analyze and characterize Alzheimer’s, but an correct and beneficial symptomatic course of activity is required. The proposed show up proposes a noteworthy learning-based strategy for diagnosing and classifying Alzheimer’s malady utilizing the DenseNet-169 and ResNet-50 CNN plans. Non-Dementia, Remarkably Mild-Dementia, Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 806 https://internationalpubls.com Sensitive Dementia, and Arrange Dementia were the four classifications of Alzheimer’s Sickness in this outline. Within the middle of the arranging and testing stages, the DenseNet- 169 strategy outmaneuvered. This proposed approach may be utilized to do real-time examination and classification of Alzheimer’s defilement. Interior long run, we organize to open up the defilement range with more information sets and utilize the specific measures to recognize the system’s accuracy. Future Scope : Creating examine underscores a incredible and multifaceted future for Alzheimer’s revelation utilizing significant learning. A basic street is multimodal integration—combining MRI, PET channels, genomic, and clinical data—with temporal modeling by implies of RNNs, LSTMs, or GNNs to progress illustrative exactness and screen ailment development effectively To combat data deficiency and inclination, methodologies like trade learning, GAN/VAE based data increment, bound together learning, and the standardization of clarified datasets are picking up unmistakable quality Adjacent these endeavors, joining sensible AI—including thought components, Grad CAM, SHAP, and LIME—has gotten to be imperative for clinician accept and appear straightforwardness Too, lightweight structures like EfficientNet B0 support course of action on edge contraptions, empowering broader get to in resource-limited settings Another promising way is the gathering and hybrid modeling approach, blending CNNs, LSTMs, and thought frameworks for higher adaptability and accuracy Other than, non-invasive progressed biomarkers—such as wearable sensors, talk and behavioral data examination, and eye tracking—are creating as beneficial early pointers. At last, finishing real-world influence will require sweeping scale, up and coming clinical endorsement through randomized trials, ethical data sharpens, standardized traditions (e.g., MINIMAR), and integration into clinical workflows to ensure fair-minded, solid, and regulatory compliant sending REFERENCES 1. Bilal et al. (2020).Bilal M, Barani M, Sabir F, Rahdar A, Kyzas GZ. Nanomaterials for the treatment and diagnosis of Alzheimer’s disease: an overview. NanoImpact. 2020;20:100251. doi: 10.1016/j.impact.2020.100251. [DOI] [Google Scholar] 2. Budhiraja & Garg (2021).Budhiraja I, Garg D. Alzheimer’s disease classification using transfer learning. International Advanced Computing Conference; Cham: Springer; 2021. pp. 73–81. [Google Scholar] 3. Butt et al. (2019).Butt AUR, Ahmad W, Ashraf R, Asif M, Cheema SA. Computer aided diagnosis (CAD) for segmentation and classification of burnt human skin. 2019 International Conference on Electrical, Communication, and Computer Engineering (ICECCE); Piscataway: IEEE; 2019. pp. 1–5. [Google Scholar] 4. Chitradevi & Prabha (2020).Chitradevi D, Prabha S. Analysis of brain sub regions using optimization techniques and deep learning method in Alzheimer disease. Applied Soft https://doi.org/10.1016/j.impact.2020.100251 https://scholar.google.com/scholar_lookup?journal=NanoImpact&title=Nanomaterials%20for%20the%20treatment%20and%20diagnosis%20of%20Alzheimer%E2%80%99s%20disease:%20an%20overview&author=M%20Bilal&author=M%20Barani&author=F%20Sabir&author=A%20Rahdar&author=GZ%20Kyzas&volume=20&publication_year=2020&pages=100251&doi=10.1016/j.impact.2020.100251& https://scholar.google.com/scholar_lookup?Budhiraja%20I,%20Garg%20D.%20Alzheimer%E2%80%99s%20disease%20classification%20using%20transfer%20learning.%20International%20Advanced%20Computing%20Conference;%20Cham:%20Springer;%202021.%20pp.%2073%E2%80%9381. https://scholar.google.com/scholar_lookup?Butt%20AUR,%20Ahmad%20W,%20Ashraf%20R,%20Asif%20M,%20Cheema%20SA.%20Computer%20aided%20diagnosis%20(CAD)%20for%20segmentation%20and%20classification%20of%20burnt%20human%20skin.%202019%20International%20Conference%20on%20Electrical,%20Communication,%20and%20Computer%20Engineering%20(ICECCE);%20Piscataway:%20IEEE;%202019.%20pp.%201%E2%80%935. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 807 https://internationalpubls.com Computing. 2020;86(4):105857. doi: 10.1016/j.asoc.2019.105857. [DOI] [Google Scholar] 5. DeTure & Dickson (2019).DeTure MA, Dickson DW. The neuropathological diagnosis of Alzheimer’s disease. Molecular Neurodegeneration. 2019;14(1):1–18. doi: 10.1186/s13024-019-0333-5. [DOI] [PMC free article] [PubMed] [Google Scholar] 6. Frozza, Lourenco & De Felice (2018).Frozza RL, Lourenco MV, De Felice FG. Challenges for Alzheimer’s disease therapy: insights from novel mechanisms beyond memory defects. Frontiers in Neuroscience. 2018;12:37. doi: 10.3389/fnins.2018.00037. [DOI] [PMC free article] [PubMed] [Google Scholar] 7. Fulton et al. (2019).Fulton LV, Dolezel D, Harrop J, Yan Y, Fulton CP. Classification of Alzheimer’s disease with and without imagery using gradient boosted machines and ResNet-50. Brain Sciences. 2019;9(9):212. doi: 10.3390/brainsci9090212. [DOI] [PMC free article] [PubMed] [Google Scholar] 8. Gaudiuso et al. (2020).Gaudiuso R, Ewusi-Annan E, Xia W, Melikechi N. Diagnosis of Alzheimer’s disease using laser-induced breakdown spectroscopy and machine learning. Spectrochimica Acta Part B: Atomic Spectroscopy. 2020;171:105931. doi: 10.1016/j.sab.2020.105931. [DOI] [PMC free article] [PubMed] [Google Scholar] 9. Kaggle (2019).Kaggle Alzheimer’s dataset (4 class of Images) 2019. https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of-images. [20 January 2022]. https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of- images 10. Khan et al. (2020).Khan FA, Butt AUR, Asif M, Ahmad W, Nawaz M, Jamjoom M, Alabdulkreem E. Computer-aided diagnosis for burnt skin images using deep convolutional neural network. Multimedia Tools and Applications. 2020;79(45):34545–34568. doi: 10.1007/s11042-020-08768-y. [DOI] [Google Scholar] https://doi.org/10.1016/j.asoc.2019.105857 https://scholar.google.com/scholar_lookup?journal=Applied%20Soft%20Computing&title=Analysis%20of%20brain%20sub%20regions%20using%20optimization%20techniques%20and%20deep%20learning%20method%20in%20Alzheimer%20disease&author=D%20Chitradevi&author=S%20Prabha&volume=86&issue=4&publication_year=2020&pages=105857&doi=10.1016/j.asoc.2019.105857& https://scholar.google.com/scholar_lookup?journal=Applied%20Soft%20Computing&title=Analysis%20of%20brain%20sub%20regions%20using%20optimization%20techniques%20and%20deep%20learning%20method%20in%20Alzheimer%20disease&author=D%20Chitradevi&author=S%20Prabha&volume=86&issue=4&publication_year=2020&pages=105857&doi=10.1016/j.asoc.2019.105857& https://doi.org/10.1186/s13024-019-0333-5 https://pmc.ncbi.nlm.nih.gov/articles/PMC6679484/ https://pubmed.ncbi.nlm.nih.gov/31375134/ https://scholar.google.com/scholar_lookup?journal=Molecular%20Neurodegeneration&title=The%20neuropathological%20diagnosis%20of%20Alzheimer%E2%80%99s%20disease&author=MA%20DeTure&author=DW%20Dickson&volume=14&issue=1&publication_year=2019&pages=1-18&pmid=31375134&doi=10.1186/s13024-019-0333-5& https://doi.org/10.3389/fnins.2018.00037 https://pmc.ncbi.nlm.nih.gov/articles/PMC5808215/ https://pubmed.ncbi.nlm.nih.gov/29467605/ https://scholar.google.com/scholar_lookup?journal=Frontiers%20in%20Neuroscience&title=Challenges%20for%20Alzheimer%E2%80%99s%20disease%20therapy:%20insights%20from%20novel%20mechanisms%20beyond%20memory%20defects&author=RL%20Frozza&author=MV%20Lourenco&author=FG%20De%20Felice&volume=12&publication_year=2018&pages=37&pmid=29467605&doi=10.3389/fnins.2018.00037& https://doi.org/10.3390/brainsci9090212 https://pmc.ncbi.nlm.nih.gov/articles/PMC6770938/ https://pubmed.ncbi.nlm.nih.gov/31443556/ https://scholar.google.com/scholar_lookup?journal=Brain%20Sciences&title=Classification%20of%20Alzheimer%E2%80%99s%20disease%20with%20and%20without%20imagery%20using%20gradient%20boosted%20machines%20and%20ResNet-50&author=LV%20Fulton&author=D%20Dolezel&author=J%20Harrop&author=Y%20Yan&author=CP%20Fulton&volume=9&issue=9&publication_year=2019&pages=212&pmid=31443556&doi=10.3390/brainsci9090212& https://doi.org/10.1016/j.sab.2020.105931 https://pmc.ncbi.nlm.nih.gov/articles/PMC8293921/ https://pubmed.ncbi.nlm.nih.gov/34295017/ https://scholar.google.com/scholar_lookup?journal=Spectrochimica%20Acta%20Part%20B:%20Atomic%20Spectroscopy&title=Diagnosis%20of%20Alzheimer%E2%80%99s%20disease%20using%20laser-induced%20breakdown%20spectroscopy%20and%20machine%20learning&author=R%20Gaudiuso&author=E%20Ewusi-Annan&author=W%20Xia&author=N%20Melikechi&volume=171&publication_year=2020&pages=105931&pmid=34295017&doi=10.1016/j.sab.2020.105931& https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of-images https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of-images https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of-images https://doi.org/10.1007/s11042-020-08768-y https://scholar.google.com/scholar_lookup?journal=Multimedia%20Tools%20and%20Applications&title=Computer-aided%20diagnosis%20for%20burnt%20skin%20images%20using%20deep%20convolutional%20neural%20network&author=FA%20Khan&author=AUR%20Butt&author=M%20Asif&author=W%20Ahmad&author=M%20Nawaz&volume=79&issue=45&publication_year=2020&pages=34545-34568&doi=10.1007/s11042-020-08768-y& https://scholar.google.com/scholar_lookup?journal=Multimedia%20Tools%20and%20Applications&title=Computer-aided%20diagnosis%20for%20burnt%20skin%20images%20using%20deep%20convolutional%20neural%20network&author=FA%20Khan&author=AUR%20Butt&author=M%20Asif&author=W%20Ahmad&author=M%20Nawaz&volume=79&issue=45&publication_year=2020&pages=34545-34568&doi=10.1007/s11042-020-08768-y&