Chest X-ray Diagnosis using CNN (AlexNet)

This project explores the application of Convolutional Neural Networks (CNNs) to automate the diagnosis of chest X-ray images, specifically classifying them into "Normal" or "Pneumonia". Leveraging the pre-trained AlexNet architecture and the Harvard Chest X-ray Dataset 2, we fine-tuned the model for high accuracy in medical image classification.

🩺 Motivation

Manual diagnosis of chest X-rays is time-consuming and prone to human error. Our objective is to assist radiologists by providing a fast and reliable AI-based diagnostic tool.

📊 Dataset

• Harvard Chest X-ray Dataset 2
• Contains labeled chest X-ray images with pathologies.

🧠 Model

• AlexNet (Pretrained on ImageNet)
• Final layer modified for binary classification (Normal vs Pneumonia).

🛠️ Methodology

1. Data Preprocessing: Resizing, normalization, data augmentation (rotation, flipping).
2. Training:
   • Optimizer: SGD / Adam
   • Loss Function: CrossEntropyLoss
   • Epochs: 25
   • Batch Size: 32
   • Learning Rate Scheduler
3. Fine-tuning on train/val/test splits.

🏆 Results & Observations

• Good generalization on unseen test data.
• Data augmentation improved performance.
• Some confusion in early-stage pneumonia due to visual similarities.

✅ Conclusion

• Demonstrated feasibility of AI-assisted radiological diagnosis.
• Transfer learning significantly boosted performance and reduced training time.

🚀 Future Scope

• Try deeper models like ResNet, DenseNet.
• Multi-label classification (e.g., TB, COVID-19).
• Real-time deployment via web/mobile interfaces.
• Explainability tools like Grad-CAM.