About The Project
Stroke Disease & Its Stages Using Deep Learning
This Flask-based platform combines machine learning, convolutional neural networks, and Grad-CAM explainability to support stroke analysis from both structured clinical data and medical images. The goal is to make the system practical, reviewable, and easy to demonstrate in an academic or healthcare setting.
Research Focus
Early support for stroke assessment
Improve diagnostic awareness by presenting risk outputs, class predictions, and interpretable evidence in one workflow.
System Strength
Hybrid intelligence
Blend numerical prediction, image classification, history, and reporting without forcing users across multiple tools.
Problem Statement
Why this system matters
Stroke diagnosis can depend on both patient attributes and imaging observations. Manual interpretation and disconnected tools may slow review or make results harder to compare. This project addresses that challenge by bringing prediction, visualization, record keeping, and reporting into a single application.
Key Objective
Provide a responsive and explainable web interface for stroke-related prediction so users can move from data input to interpretable output with minimal friction.
Objectives
What the platform is built to achieve
Clinical Risk Estimation
Use tabular ML to assess stroke likelihood from demographic and clinical values.
Imaging Classification
Analyze medical images through a CNN model and identify supported stroke classes.
Explainable AI
Attach Grad-CAM heatmaps so image predictions include meaningful visual interpretation.
Data Continuity
Maintain user history and reports to support review, tracking, and presentation.
Proposed Workflow
From secure entry to report generation
1. User Authentication
Users register or log in to access protected prediction and record-management features.
2. Model Execution
The system routes users to numerical prediction or image prediction depending on the input type.
3. Explainable Results
Predictions are shown alongside Grad-CAM visual interpretation for image-based inference.
4. Save, History, and Reports
Records are stored in SQLite for later review in history and summary reports.
Tech Stack
Technologies behind the interface and models
Frontend
Flask templates, Tailwind CSS, responsive cards, and local SVG illustrations.
Backend Integration
Flask routes coordinate user flow, prediction endpoints, and report rendering.
Machine Learning
A tabular ML model supports numerical stroke-risk estimation from structured features.
Deep Learning
A CNN model performs image-based classification and visual reasoning support with Grad-CAM.
Modules & Features
Functional areas available in the application
Authentication Module
Handles user registration, login, protected access, and logout for a personalized workflow.
Numerical Prediction Module
Collects clinical inputs and returns model-driven risk predictions for stroke analysis.
Image Prediction Module
Accepts CT or MRI images, runs CNN-based classification, and displays Grad-CAM outputs.
History & Report Module
Lets users revisit prior predictions, track stored outcomes, and view report summaries.
Explainability
Grad-CAM helps users trust the image model
Grad-CAM highlights the image regions that influenced the CNN prediction most strongly. In this project, it adds interpretability to the imaging pipeline and supports presentation-ready reasoning for reports, demos, and academic explanation.
Why it matters
Explainable visual evidence makes model outputs easier to verify, discuss, and compare.
Database, History, and Reports
Structured persistence for every important result
SQLite stores user-linked prediction details so the application can present a reliable history and summary view. This supports project evaluation, result recall, and quick reporting without forcing users to re-run earlier cases.
History
Track saved numerical and imaging predictions over time.
Report
Generate an organized summary layer for review, charts, and demonstrations.