Pneumonia X-ray Detector (CNN)

A deep learning project to detect Pneumonia from chest X-ray images using a Convolutional Neural Network (CNN).

This project has:

• Frontend (HTML, CSS, JS) → inside the frontend folder for remote access.
• Backend API (Flask) → for predictions.
• Streamlit Deployment → a standalone web app for easy usage without dealing with CORS issues.

---

Live Demo

• Streamlit Deployment: https://czgk8ceahwswtw2wmd97x3.streamlit.app
  (Click to open the deployed app and try it live)

• Frontend + Backend Setup:

  • Frontend folder: /frontend
  • Backend served on: https://pneumonia-x-ray-detector-cnn.onrender.com

---

Dataset

The dataset used is from Kaggle:
Chest X-Ray Images (Pneumonia)

Dataset Structure:

dataset/ │ ├── train/ │ ├── NORMAL/ │ └── PNEUMONIA/ │ ├── test/ │ ├── NORMAL/ │ └── PNEUMONIA/ │ └── val/ ├── NORMAL/ └── PNEUMONIA/

• NORMAL → X-rays without pneumonia.
• PNEUMONIA → includes bacterial and viral pneumonia images.

---

🧠 CNN Model Architecture

We used a custom CNN with 5 convolutional layers.
The model was trained using TensorFlow and Keras.

Layer Type	Filters / Units	Activation	Notes
Conv2D	32	ReLU	Input shape (150, 150, 1)
MaxPooling2D	2x2	-	Downsampling
Conv2D	64	ReLU	+ Dropout (0.1)
MaxPooling2D	2x2	-	Downsampling
Conv2D	64	ReLU	-
MaxPooling2D	2x2	-	Downsampling
Conv2D	128	ReLU	+ Dropout (0.2)
MaxPooling2D	2x2	-	Downsampling
Conv2D	256	ReLU	+ Dropout (0.2)
MaxPooling2D	2x2	-	Final pooling
Flatten	-	-	Flatten output
Dense	128	ReLU	Fully connected
Dropout	0.2	-	Prevent overfitting
Dense (Output)	1	Sigmoid	Binary classification

---

Model Performance

The model was trained for 12 epochs with data augmentation applied to prevent overfitting.

Metric	Value
Accuracy	92.46%
Loss	0.27

---

How It Works

1. Input: User uploads a chest X-ray image.
2. Preprocessing:
   • Image is converted to grayscale.
   • Resized to 150x150 pixels.
   • Normalized to the range [0, 1].
3. Prediction: The trained CNN model classifies the image as:
   • NORMAL
   • PNEUMONIA
4. Output: Displays the prediction and confidence score.

---

Project Flow

1. Frontend (HTML/CSS/JS)

   • Simple interface for uploading chest X-ray images.
   • Sends the image to the backend API for prediction.

2. Backend (Flask)

   • Handles requests from the frontend.
   • Runs the trained CNN model (pneumonia_model.h5).

3. Streamlit App

   • Standalone web app combining frontend and backend.
   • Deployed on Streamlit Cloud for easy sharing.

---

Deployment

Frontend + Backend:

• Backend: Render (Flask + Gunicorn)
• Frontend: Vercel (Static HTML, CSS, JS)

Streamlit Deployment:

• Host: Streamlit Community Cloud
• Link: https://your-app-name.streamlit.app

---

Screenshots

Below are placeholders for four images. Replace them with actual screenshots later.

Sample image(normal)	Sample image(pneumonia)	Training vs Validation	Data Distribution

---

Tech Stack

• Python (TensorFlow, Keras, NumPy, Pillow)
• Flask (Backend API)
• Streamlit (Final Deployment)
• HTML, CSS, JS (Frontend UI)
• Render (Backend hosting)
• Vercel (Frontend hosting)

---

Running Locally

1. Clone the repo:

   git clone https://github.com/<your-username>/Pneumonia-X-ray-Detector-CNN.git

2. Navigate into the project folder:

   cd Pneumonia-X-ray-Detector-CNN

3. Install dependencies:

   pip install -r requirements.txt
   

4. Run the Streamlit app:

   streamlit run streamlit_app.py
   

Author

Arjjun S

Building solutions in deep learning and computer vision with a focus on real-world applications.