🧠 Brain Tumor Classification — Medical Imaging MLOps Pipeline

An end-to-end MLOps pipeline that classifies brain MRI scans into diagnostic categories using Deep Learning (PyTorch), packaged as decoupled microservices in Docker containers, and deployed live on AWS cloud infrastructure.

---

📸 Demo

Live deployment on AWS EC2 — MRI scan input → real-time diagnostic classification

---

🏗️ System Architecture

+---------------------------------+
|         Streamlit Web GUI       |
|    (Port 8501 - User Interface) |
+---------------------------------+
                |
                | HTTP POST /predict (Image Payload)
                v
+---------------------------------+
|          FastAPI Backend        |
|     (Port 8000 - Model API)     |
+---------------------------------+
                |
                | Inference Tensor Processing
                v
+---------------------------------+
|        PyTorch CNN Model        |
|  (brain_tumor_model.pth Weights)|
+---------------------------------+

1. Frontend (Streamlit): Web interface for uploading MRI scans and triggering real-time diagnostic inference.
2. Backend (FastAPI): Asynchronous API that preprocesses input images, runs tensor inference via PyTorch, and returns structured JSON predictions.
3. Containerization & Deployment: Multi-container setup managed via Docker Compose, deployed on AWS EC2 (Ubuntu, eu-north-1).

---

📊 Model Performance

• Task: Multi-class brain MRI classification — Glioma / Meningioma / Pituitary Tumor / No Tumor
• Framework: PyTorch CNN
• Deployment: Live REST API inference — image upload → JSON prediction → UI render

---

🛠️ Tech Stack

Layer	Technology
Machine Learning	PyTorch, Torchvision, Pillow, NumPy
API	FastAPI, Uvicorn
Frontend	Streamlit
Containerization	Docker, Docker Compose V2
Cloud	AWS EC2 (Ubuntu Linux, eu-north-1)
Version Control	Git, GitHub

---

📁 Repository Structure

medical-imaging-mlops/
│
├── models/                      # Model weights (excluded via .gitignore)
│   └── brain_tumor_model.pth    # Download link — see below
├── src/
│   ├── api.py                   # FastAPI inference endpoints
│   └── app.py                   # Streamlit web application
├── Dockerfile.backend           # Docker image — FastAPI backend
├── Dockerfile.frontend          # Docker image — Streamlit frontend
├── docker-compose.yml           # Container orchestration
├── requirements.txt             # Python dependencies
└── README.md

Model weights: Available on request or via (Google Drive Link) — place in models/ before running.

---

🚀 How to Run

Local (Docker required)

git clone https://github.com/Sachinsara14/medical-imaging-mlops
cd medical-imaging-mlops
# Add brain_tumor_model.pth to models/ folder
docker compose up --build

Open http://localhost:8501 in your browser.

AWS Deployment

Deployed on EC2 using Docker Compose V2. Access the live instance at: [your EC2 public IP]:8501

---

🔬 Part of a Larger Medical Imaging Portfolio

This project is one component of an ongoing medical imaging ML portfolio:

• Brain Tumor Classification (this repo) — MLOps deployment
• Cardiac Histopathology Segmentation — Attention UNet, active learning, WSI analysis (6,627 images, Dice 0.8956)

---

👤 Author

Sachinkumar P — M.Tech., IIT Roorkee
github.com/sachinkumarp-code | LinkedIn