| Category | Description |
|---|---|
| Domain | Computer Vision, Machine Learning |
| Subdomain | Deep Learning, Image Classification |
| Techniques | CNN, Transfer Learning (InceptionV3, ImageNet) |
| Application | Medical Imaging, Disease Diagnosis, Radiology |
This project aims to detect Pneumonia from chest X-ray images using both a custom-built Convolutional Neural Network (CNN) and a fine-tuned InceptionV3 model.
- Utilized a dataset of 5,856 labeled X-ray images (1.15 GB) categorized into two classes: Pneumonia and Normal.
- For transfer learning:
- Removed the top layers of the pre-trained InceptionV3 model.
- Froze the early convolutional blocks.
- Fine-tuned deeper layers on the new dataset.
- The custom CNN achieved:
- Accuracy: 89.53% on test set
- Loss: 0.41
| Category | Tools / Libraries |
|---|---|
| Language | Python |
| Environment | Anaconda |
| Libraries | Keras, TensorFlow, InceptionV3, ImageNet |
Dataset Name: Chest X-Ray Images (Pneumonia)
| Set | Images | Size |
|---|---|---|
| Training | 5,216 | ~1.07 GB |
| Validation | 320 | ~42.8 MB |
| Testing | 320 | ~35.4 MB |
| Total | 5,856 | ~1.15 GB |
- Number of Classes: 2 (Pneumonia, Normal)
- Batch Size: 64
- Epochs: 30
- Training Time: ~2 Hours
- Optimizer: Adam
- Loss Function: Categorical Crossentropy
- Base model pre-trained on ImageNet.
- Top layers removed and replaced with custom dense layers.
- Lower layers frozen; deeper layers fine-tuned for medical image specificity.
| Metric | Value |
|---|---|
| Accuracy (F1) | 89.53% |
| Loss | 0.41 |
| Precision | 88.37% |
| Recall (Pneumonia) | 95.48% |
🔍 The high recall for Pneumonia ensures minimal false negatives — a critical requirement in medical diagnostics.
This project highlights the effectiveness of deep learning in medical imaging, especially in life-critical applications like pneumonia detection. While the custom CNN provided strong baseline results, transfer learning with InceptionV3 further accelerated model convergence and enhanced generalization.