Abstract
Alzheimer’s disease is a progressive neurodegenerative disorder that gradually impairs memory, cognition, and functional ability. Timely detection plays a crucial role in clinical decision-making, yet traditional diagnosis depends on manual MRI interpretation, which is subjective, slow, and requires expert radiological skills. This project proposes an automated Alzheimer detection framework using a Convolutional Neural Network (CNN) trained on MRI brain images. The system preprocesses input images, extracts discriminative spatial features, and classifies them into Normal or Alzheimer-affected categories. The trained model is deployed through two user-friendly interfaces: a Flask-based web application for online screening and a Tkinter desktop GUI for offline use. Both environments ensure consistent inference, rapid prediction, and simple interaction suitable for clinical and non-clinical contexts. The proposed approach increases diagnostic accuracy, reduces dependence on manual analysis, and improves accessibility for early-stage Alzheimer identification.
Keywords: Alzheimer’s Disease, Deep Learning, Convolutional Neural Network, MRI Classification, Flask Application, Tkinter GUI, Medical Image Analysis, Automated Diagnosis.












Introduction
Alzheimer’s disease (AD) is the most common cause of dementia and is characterised by progressive synaptic loss, accumulation of amyloid-β plaques and tau tangles, and measurable structural and functional brain changes long before clinical symptoms become disabling. Structural MRI captures macroscopic patterns of atrophy — notably hippocampal shrinkage, medial temporal lobe degeneration and cortical thinning — that correlate with disease stage and cognitive decline. Modern neuroimaging studies therefore treat MRI-derived features as essential biomarkers for early detection and prognosis. Large multicentre longi tudinal repositories such as the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and OASIS provide standardized MRI volumes, clinical scores and follow-up labels; these datasets underpin most contemporary AI studies and enable comparisons across methods.
Recent years have seen deep learning — especially convolutional neural networks (CNNs) — supplant handcrafted feature pipelines in MRI-based AD classification. Surveys of the field show consistent themes: (1) 3D volumetric models can exploit full spatial context but require more data and compute; (2) slice-based 2D CNNs or hybrid CNN+sequence models (e.g., CNN features passed to RNN/transformer layers) offer strong trade-offs between accuracy and practicality; and (3) transfer learning and careful data augmentation are critical when labeled medical datasets are limited. Reviews also emphasize that multi-modal inputs (MRI + PET + clinical scores) typically yield better clinical performance than single-modality models, though they raise data-harmonization and missing-modality challenges.

Objectives
The major objectives of the project are:
1. To design and develop a deep learning–based Alzheimer detection system using MRI images.
2. To construct a CNN model capable of extracting spatial features and performing binary classification.
3. To implement preprocessing techniques that normalize and standardize MRI inputs.
4. To deploy the trained model through a Flask web application with secure authentication and image upload facilities.
5. To develop a Tkinter-based desktop application enabling offline Alzheimer detection.
6. To evaluate the system using accuracy, loss, confusion matrix, sensitivity, and specificity.
7. To ensure the framework is scalable, interpretable, and user-friendly for diverse use environments.