🧠 CT Scan Image Denoising & Brain Tumor Analysis

📖 Introduction

This project focuses on enhancing brain CT scans by reducing acquisition noise using a CNN-based autoencoder, followed by tumor detection on the refined images.

The workflow ensures that critical anatomical details are preserved while improving diagnostic accuracy.
A lightweight Flask web app allows clinicians or researchers to upload CT scans (.dcm format) and instantly visualize denoised results along with tumor classification.

Deployed seamlessly on AWS EC2 (remote access via PuTTY).

---

📂 Project Layout

CT-Image-Denoising/
├── models/             # Saved deep learning models (autoencoder + classifier)
├── static/             # Assets for the Flask frontend (CSS, images, JS)
├── templates/          # HTML templates for UI (upload page, result page)
├── app.py              # Core pipeline: preprocessing, inference, evaluation
├── requirements.txt    # List of Python dependencies
└── README.md           # Documentation

👉 app.py acts as the entry point — handling:

• DICOM loading & preprocessing
• Autoencoder inference (denoising)
• Tumor classification on denoised scans
• Evaluation metrics (SNR, classification report)
• Flask-based web serving

---

🚀 Core Features

• 🧠 Noise Reduction: Autoencoder removes CT noise while retaining diagnostic details.
• 🩺 Tumor Prediction: Classifier identifies tumor presence on enhanced images.
• 📊 Metrics: Includes SNR improvement tracking & classification reports.
• 🌍 Cloud Deployment: Flask app hosted on AWS EC2 for remote usage.

---

📊 Performance Snapshot

✅ Classification Accuracy:

• Before denoising → 0.37
• After denoising → 0.84

✅ Signal-to-Noise Ratio (SNR):

Condition	SNR (dB)
Raw CT (noisy)	2.94
After Denoising	15.58

---

🖼️ Visual Results

🔹 CT Denoising Example

Noise vs. Enhanced Image

---

⚡ Getting Started

🔧 Local Setup

1. Clone the repo & install dependencies:

   pip install -r requirements.txt

2. Add trained weights (.h5 / .pt) into the models/ folder.
3. Run the app:

   python app.py

4. Open http://127.0.0.1:5000/ in your browser → Upload a .dcm scan → View results.

---

👨‍💻 Author

📌 Developed by: Shubham Vishwakarma
💬 Feel free to reach out for collaboration or research discussions.

---

✨ In short: This system transforms noisy CT scans into clinically useful images, leading to better tumor detection and higher diagnostic confidence.