NeuroSage

Starting Screen
MRI uploaded
Analysis Done (using Grad-Cam and CNN)
Word Test starts
Words are flashing
Word Recall
Score provided
Whack-A-Mole Test
Mole Moving
Cognitive Clock Test
Final Results Computation

Project Inspiration & Overview

The main inspiration for this project was that it was a DO GOOD initiative. We aimed to build it as holistically as possible to extract maximum learning value across domains.

This project served as a strong introduction to:

Web development
Using Grad-CAM for the first time
End-to-end frontend development, having previously worked mostly on backend systems

We built a CNN model from scratch, and analyzing MRI scans in conjunction with different cognitive test scores proved to be a highly valuable learning experience.

Dataset & Model Development

We began by identifying a suitable dataset and used the following Kaggle resource:
https://www.kaggle.com/datasets/uraninjo/augmented-alzheimer-mri-dataset

Steps followed:

Trained a CNN model on the dataset
Used Grad-CAM to generate heatmaps and final overlays for interpretability
Classified MRI scans into the following categories:
- MildDemented
- ModerateDemented
- NonDemented
- VeryMildDemented

Understanding Clinical Diagnosis

To understand how Alzheimer’s is diagnosed cognitively by clinicians, we studied the following references:

Based on this research, we designed three major cognitive tests.

Cognitive Tests Designed

A) Word Recall Test

We built a simplified version of real-world word recall tests used by doctors.

Test Flow:

5 words are shown from a diverse word bank
Words are flashed at a speed of 1.2 seconds
The user is asked to recall the words one by one

Metrics Evaluated:

Accuracy
Speed

Reference:
https://www.sciencedirect.com/science/article/abs/pii/027826269090047R

B) Whack-A-Mole Reaction Test

The above research highlighted the importance of reaction time in cognitive assessment. To make this both engaging and effective, we implemented a Whack-A-Mole game—a familiar childhood concept.

Entire UI/UX was designed and developed by us from scratch

Metrics Evaluated:

Mean Reaction Time
Accuracy

C) Clock Drawing Test (Developer-Friendly Version)

We created a simplified, developer-friendly version of the traditional Clock Drawing Test.

Objective:

Evaluate how accurately the user understands a target time
Measure how well they can translate it onto a user-controlled clock interface

Final Diagnosis Logic

In the final step:

MRI embeddings and cognitive test scores are combined
Dynamic weights are applied to each component
A simplified final score is computed

This score helps in indicating the potential need for professional medical consultation, rather than providing a definitive diagnosis.

Conclusion

This project integrates AI-based medical imaging, human–computer interaction, and cognitive science into a single diagnostic support framework, maximizing both social impact and technical learning.