Cognify

Final Year Project

Student Name: Josh Crotty
Student Number: 20096881
Academic Title: Cloud-Based System for detection of Alzheimer’s Disease using Deep Learning on MRI Images
Supervisor: Dr. Bernard Butler

Project Description

Cognify is a cloud-based AI system designed to analyse MRI images and detect signs of Alzheimer’s Disease (AD) using deep learning techniques. This system leverages Magnetic Resonance Imaging (MRI) to provide a reliable and efficient diagnostic tool for researchers and healthcare professionals. To ensure optimal performance, multiple deep learning models are trained and compared. A variety of hyperparameter tuning strategies are explored to improve accuracy and robustness. The models are evaluated using key performance metrics such as accuracy, sensitivity (recall), specificity, AUC-ROC, and F1-score. This project will focus on the comparison of selected models along with the effects of augmentation and transfer learning, compiling, and analysing each of the metrics to determine an effective architecture for Alzheimer’s Disease classification.

Additionally, sample MRI images from the ADNI dataset are showcased below to illustrate the type of data used in the project. The final report detailing the methodology, results, and findings is available below.

Sample scan set

MRI Sample
Unprocessed middle slice from each label in the axial plane. (CN: cognitively normal; MCI: mild cognitive impairment; AD: Alzheimer’s disease)

Prototype model loss and accuracy curves

Figures below show the prototype model ResNet-50 training and testing loss curves along with test accuracy over 20 epochs.

Training and Test Loss Curve
Figure 1: Train and Test Loss Curves
Model Accuracy Curve
Figure 2: Test Accuracy

Technologies

Category Technology Used
Frontend React, Material UI, NiiVue
Backend Node.js, Express.js
Database MongoDB
Authentication JWT
Deep Learning PyTorch, MONAI
Model Deployment Streamlit, Docker, TorchServe, AWS
Data Processing NumPy, Pandas, FSL (FLIRT/BET), Nipype
Documentation & Writing LaTeX, Overleaf, Zotero, Matplotlib

Project Report

🔗 Click here to download the full report (PDF)