OncoScan AI is a full-stack explainable artificial intelligence system developed for detecting brain tumors from MRI images using a Convolutional Neural Network (CNN). The system integrates Grad-CAM for heatmap-based explainable AI visualization, a FastAPI backend for prediction and data management, and a web-based frontend interface for user interaction.

The platform also includes PDF medical report generation, prediction history tracking, and an admin analytics dashboard. The frontend is deployed on Netlify and the backend is deployed on Render. The trained deep learning model is optimized and deployed using ONNX Runtime for efficient inference.

This project integrates Machine Learning, Backend Development, Frontend Development, Database Management, REST API Development, and Cloud Deployment into a complete AI-based web platform.

• Brain tumor detection from MRI images
• Explainable AI using Grad-CAM heatmaps
• PDF medical report generation
• User authentication system
• Prediction history tracking
• Admin dashboard with analytics
• Charts and statistics visualization
• Cloud deployment (Netlify + Render)
• ONNX optimized model inference
• Full-stack AI web application

Brain tumor diagnosis using MRI images is a complex and time-consuming process that requires expert radiologists. The goal of this project is to assist medical professionals by developing an AI system that can automatically analyze MRI images and predict whether a tumor is present or not.

The system also provides explainable AI visualization (Grad-CAM heatmap) to show which region of the MRI image influenced the model’s decision.

The brain MRI dataset used for training the model was collected from publicly available medical imaging datasets and brain MRI tumor classification datasets available on Kaggle and other open research sources.

The dataset contains MRI brain images categorized into:

• Tumor
• No Tumor

oncoscan-ai/
│
├── backend/ # FastAPI backend + AI integration
│ ├── main.py # Main FastAPI application
│ ├── database.py # Database connection & queries
│ └── pycache/
│
├── frontend/ # Frontend (Netlify hosted)
│ ├── admin/
│ │ ├── dashboard.html
│ │ ├── dashboard.js
│ │ ├── stats.html
│ │ └── stats.js
│ │
│ ├── user/
│ │ ├── history.html
│ │ └── history.js
│ │
│ ├── index.html # Upload page
│ ├── login.html # Login page
│ ├── result.html # Result page
│ ├── script.js
│ ├── style.css
│ ├── config.js # Backend API URL config
│ └── _redirects # Netlify redirects
│
├── models/ # Machine learning models
│ ├── brain_tumor_model.h5
│ ├── model_fixed.h5
│ ├── model_fixed.keras
│ └── model.onnx # Optimized ONNX model
│
├── Images/ # README screenshots
│
├── static/ # Uploaded images, heatmaps, PDFs
│ ├── images/
│ ├── heatmaps/
│ └── reports/
│
├── utils/
├── tests/
├── notebooks/
├── results/
│
├── fix_model.py
├── database.db
├── requirements.txt
├── runtime.txt
├── Procfile
├── .gitignore
└── README.md

The brain tumor detection model is built using Deep Learning with Convolutional Neural Networks (CNN).

• Model Type: Convolutional Neural Network (CNN)
• Task: Binary Image Classification
• Classes: Tumor / No Tumor
• Framework: TensorFlow / Keras
• Deployment Model: ONNX Runtime
• Image Processing: OpenCV
• Output:
  • Prediction (Tumor / No Tumor)
  • Confidence Score

Grad-CAM (Gradient-weighted Class Activation Mapping) is used to visualize which region of the MRI image influenced the model’s prediction.

Input Image
↓
CNN Forward Pass
↓
Compute Gradients
↓
Generate Heatmap
↓
Overlay Heatmap on Image

This makes the AI system explainable and suitable for medical applications.

The backend of the system is developed using FastAPI.

• User Authentication
• Image Upload Handling
• AI Model Prediction
• Heatmap Generation
• PDF Report Generation
• Database Storage
• Prediction History
• Admin Dashboard APIs
• Statistics and Analytics
• Record Deletion

{
  "prediction": "Tumor",
  "confidence": 97.94,
  "heatmap": "static/heatmaps/heatmap_12.png",
  "report": "static/reports/report_12.pdf"
}

The system uses SQLite database to store user accounts and prediction records.

One User → Many Predictions

The frontend of the system is built using HTML, CSS, and JavaScript.

• User Registration
• User Login
• Upload MRI Image
• Drag and Drop Image Upload
• Image Preview
• Tumor Prediction Result
• Confidence Score Display
• Heatmap Visualization
• PDF Report Download
• Prediction History

• Admin Dashboard
• Total Predictions
• Tumor Cases Count
• No Tumor Cases Count
• Total Patients
• Recent Predictions Table
• Delete Prediction Records
• Analytics Charts

Charts are implemented using Chart.js.

Charts included:

• Prediction Distribution (Pie Chart)
• Daily Prediction Activity (Line Chart)

User (Browser) | v Frontend (HTML, CSS, JavaScript) | v Netlify (Frontend Hosting) | v FastAPI Backend (Render) | v ONNX Runtime | v AI Model (model.onnx) | v Database (SQLite)

The trained TensorFlow/Keras model was converted to ONNX format to improve deployment performance and reduce dependency issues on cloud platforms like Render.

Benefits of ONNX:

Faster inference Lower memory usage No TensorFlow dependency on server Better deployment compatibility Smaller runtime environment

git clone https://github.com/masud744/oncoscan-ai cd oncoscan-ai python -m venv venv venv\Scripts\activate pip install -r requirements.txt uvicorn backend.main:app --reload

Create a .env file: DATABASE_URL=sqlite:///database.db MODEL_PATH=models/model.onnx

User Login ↓ Upload MRI Image ↓ Image Preprocessing ↓ CNN Model Prediction ↓ Generate Heatmap ↓ Save Data to Database ↓ Generate PDF Report ↓ Show Result to User ↓ Admin Dashboard Analytics

• TensorFlow
• Keras
• CNN
• Grad-CAM
• OpenCV
• ONNX Runtime
• NumPy

• FastAPI
• Python
• SQLite
• ReportLab
• Uvicorn

• HTML
• CSS
• JavaScript
• Chart.js

• Netlify
• Render
• GitHub

• Password hashing
• Email notification system
• Multiple image prediction
• Cloud image storage
• Model accuracy dashboard
• Docker deployment
• JWT authentication
• Role-based access control
• Mobile responsive UI

Shahriar Alom Masud
B.Sc. Engg. in IoT & Robotics Engineering
University of Frontier Technology, Bangladesh
Email: shahriar0002@std.uftb.ac.bd
LinkedIn: https://www.linkedin.com/in/shahriar-alom-masud

This project is licensed under the MIT License.

If you like this project, give it a star on GitHub!