Dysgraphia Detection Web Platform

Reference

Publication

This project is published as:

• Dysgraphia Detection Using Deep Learning Techniques (IEEE Xplore)

Project Summary

Overview

Dysgraphia Detection Web App is a full-stack platform (React, Flask, TensorFlow, PyTorch, Gemini) developed from Jan 2025 to Mar 2025. The application enables users to upload handwriting samples for dysgraphia diagnosis and receive clear, structured reports.

Key Features and Results

• The model was trained on 249 handwriting images and tested on 50+ real entries.
• A custom CNN-BiLSTM architecture achieved 91% accuracy in handwriting classification.
• The dataset was expanded with 900+ GAN-generated samples to improve robustness.
• An integrated LLM (Large Language Model) detects dysgraphia symptoms, improving interpretability by 30%.

This advanced web application provides automated dysgraphia screening using deep learning and generative AI. The stack includes Next.js, TypeScript, Tailwind CSS, and a Python Flask API.

Features

• Upload handwriting samples and receive instant dysgraphia risk analysis
• ML-powered backend (Keras/TensorFlow) for classification
• Google Gemini AI for feature extraction and natural language feedback
• Secure authentication and user management
• Modern, responsive UI

Screenshots

Home Page

Testing for Potential Dysgraphia

Testing for Low Potential Dysgraphia

Results of Various Models Trained to Classify

Proposed Architecture of the Model

Getting Started

Clone the repo and install dependencies:

git clone https://github.com/NikhilKartha5/Dysgraphia-Detection.git
cd Dysgraphia-Detection
npm install

Start the development server:

npm run dev
# or
yarn dev
# or
pnpm dev

Open http://localhost:3000 in your browser.

API

The Flask backend exposes /api/predict for ML-powered handwriting analysis. See api/app.py for details.

Tech Stack

• Next.js, React, TypeScript, Tailwind CSS
• Python, Flask, TensorFlow, Google Gemini AI

License

Released under the MIT License. See LICENSE for details.

Open http://localhost:3000 with your browser to see the result.