Pneumonia-Disease-Detection

🧠 Pneumonia Disease Detection using MobileNetV2 Model (Deep Learning)

This project presents an AI-powered pneumonia detection system using MobileNetV2, a lightweight convolutional neural network, to classify Normal, Bacterial Pneumonia, and Viral Pneumonia from chest X-ray images. The model integrates image preprocessing (CLAHE and Median Filtering) to improve feature extraction and diagnostic accuracy. The final trained model is converted to TensorFlow Lite for real-time deployment on Android mobile devices.

---

🩺 Abstract

Pneumonia is a respiratory disease that causes inflammation in the lungs. Early detection is crucial for effective treatment. This system automates pneumonia detection using deep learning to reduce manual diagnosis time. MobileNetV2 provides high accuracy, low computation cost, and mobile deployment capability, making it ideal for real-time medical diagnostics.

---

🚀 Objectives

• Automate pneumonia detection using deep learning.
• Compare model accuracy with and without preprocessing.
• Deploy the trained model to a mobile application using TensorFlow Lite.
• Reduce diagnosis time and improve accessibility.

---

🛠️ Technologies Used

Category	Technology
Language	Python
Framework	TensorFlow, Keras
Environment	Google Colab
Deployment	TensorFlow Lite, Android Studio
Libraries	NumPy, Pandas, Scikit-learn, OpenCV, Matplotlib
Visualization	Seaborn, Matplotlib
Storage	Google Drive
OS	Windows 11

---

🧩 System Modules

1. Dataset Preparation

• Dataset includes 7927 chest X-ray images categorized into Normal, Bacterial Pneumonia, and Viral Pneumonia.
• Preprocessing includes CLAHE (Contrast Limited Adaptive Histogram Equalization) for contrast enhancement and Median Filtering for noise removal.
• Dataset split: 70% Train, 20% Validation, 10% Test.
• Evaluated preprocessing using PSNR (Peak Signal-to-Noise Ratio) with an average of 48.86 dB.

2. Model Development

• Used MobileNetV2 with transfer learning from ImageNet weights.
• Input image size: 128x128 pixels.

• Optimizer: Adam

  Loss: Categorical Crossentropy

  Epochs: 25

  Batch size: 32.

3. Model Accuracy Comparison

| Configuration | Train Acc | Validation Acc | Test Acc | |—————-|————|—————-|———–| | Without Preprocessing | 94% | 84% | 83% | | With Preprocessing | 99% | 86% | 84% |

4. Performance Metrics

| Metric | Without Preprocessing | With Preprocessing | |———|————————|——————–| | Precision | 83% | 84% | | Recall | 83% | 84% | | F1-Score | 83% | 84% |

---

📱 Mobile Application Integration

• Model converted to .tflite format for deployment.
• Integrated into Android Studio with TensorFlow Lite Interpreter.
• Allows users to upload X-ray images and view predictions instantly.

App Features:

• Predicts: Normal, Bacterial Pneumonia, Viral Pneumonia.
• Lightweight, real-time, and mobile-optimized model.

---

🧠 Model Workflow

1. Data Collection
   • Gathered chest X-ray images labeled as Normal, Bacterial Pneumonia, and Viral Pneumonia.
2. Image Preprocessing
   • Applied CLAHE for contrast enhancement.
   • Used Median Filtering to remove noise.
   • Resized images to 128×128 pixels.
3. Data Augmentation
   • Performed rotation, flipping, zoom, and normalization to improve model generalization.
4. Model Training (MobileNetV2)
   • Transfer learning with pretrained ImageNet weights.

   • Optimizer: Adam

     Epochs: 25

     Batch size: 32.

   • Output classes: Normal, Bacterial, Viral.
5. Model Evaluation
   • Metrics: Accuracy, Precision, Recall, F1-Score, Confusion Matrix.
6. Model Conversion
   • Converted the trained model (.h5) to TensorFlow Lite (.tflite) for mobile deployment.
7. Mobile App Integration
   • Integrated with Android Studio using TensorFlow Lite Interpreter.
   • User uploads X-ray → Model predicts disease type in real-time.
8. Output
   • ✅ Normal
   • 🦠 Bacterial Pneumonia
   • 🧬 Viral Pneumonia

---

🧪 Results

• Average PSNR: 48.86 dB – excellent image clarity after preprocessing.
• Accuracy Improvement: +5% with preprocessing.
• Model Size: Optimized for mobile deployment.
• Prediction Speed: ~2 seconds per image.

---

🌟 Advantages

• Lightweight, mobile-optimized deep learning model.
• High accuracy and real-time processing.
• Automated, non-invasive pneumonia detection.
• Works offline with no data upload required.
• Cost-effective diagnostic tool for remote areas.

---

🔮 Future Enhancements

• Integrate with cloud (Firebase / AWS) for online diagnosis.
• Include additional diseases like COVID-19 and Tuberculosis.
• Implement Explainable AI using Grad-CAM visualization.
• Add multilingual and voice support.
• Expand into full telemedicine ecosystem.

---

🧾 Conclusion

This project successfully demonstrates how MobileNetV2 can be leveraged for medical image classification with excellent performance and efficiency. The system provides a scalable, mobile-friendly solution for early pneumonia detection, improving access to diagnostic tools in underdeveloped and rural healthcare systems.

---

👨‍💻 Developer

Ravin Raj S (23322009)
Department of Computer Science and Applications
The Gandhigram Rural Institute (Deemed to be University)
📅 April 2025

---

📚 References

• Sandler, M. et al. (2018). MobileNetV2: Inverted Residuals and Linear Bottlenecks. arXiv:1801.04381
• Mooney, P. (2018). Chest X-Ray Images (Pneumonia) – Kaggle Dataset
• TensorFlow Lite: tensorflow.org/lite
• OpenCV CLAHE Docs: docs.opencv.org
• Scikit-Image Median Filter: scikit-image.org

---

🪪 License

This project is open-source and licensed under the MIT License.