Automated Pneumonia Screening via X-Ray Imaging and Deep Neural Networks #
Introduction #
- Pneumonia is a severe lung infection affecting all age groups.
- Diagnosis with X-ray is often manual and error-prone.
- Deep Learning, especially CNNs, allows for automated, high-accuracy detection.
- This project proposes an AI system to detect pneumonia using CNNs and chest X-rays.
Literature Review #
| # | Paper Title | Author(s) | Problem | Concept | Limitations |
|---|---|---|---|---|---|
| 1 | Bayesian CNN Estimation | Fernandes et al. | Limited pediatric X-ray data | Bayesian CNN | Scarce data |
| 2 | Transfer Learning from Pneumonia to COVID-19 | Hongen Lu et al. | Few public COVID datasets | Transfer Learning | Dispersed data |
| 3 | Pneumonia & COVID-19 Detection | Militante et al. | Multi-class detection | CNN (6 models) | Computation cost |
| 4 | Adaptive Deep Learning CNNs | Dionisio et al. | Pneumonia class detection | Ensemble CNN | Complexity |
| 5 | COVID-19 & Pneumonia Segmentation | Jahid Hasan et al. | Need for segmentation | DL + Segmentation | Needs annotated data |
| 6 | ImageNet Classification | Krizhevsky et al. | Image classification | AlexNet | Training cost |
| 7 | VGG for Recognition | Simonyan & Zisserman | Improve depth | VGG16 | Large model size |
| 8 | Grad-CAM Visualization | Selvaraju et al. | Interpretability | Grad-CAM | Post-hoc only |
| 9 | Early CNN for Digits | LeCun et al. | ZIP digit detection | Basic CNN | Outdated design |
| 10 | Ensemble X-ray Classifier | Dionisio et al. | Detect multiple types | Classifier Ensemble | Needs tuning |
Software & Hardware Requirements #
Software #
- Windows 8 or higher
- Python, Virtualenv
- Google Colab, TensorFlow 2.7.0
- Visual Studio Code
- CUDA 11.5, cuDNN 8.3
Hardware #
- 8GB RAM minimum
- Intel i5 / Ryzen 5
- NVIDIA GTX 1650
- 256GB SSD + 1TB HDD
⚙️ Implementation Steps #
1. Data Collection #
- Dataset: Kaggle chest X-rays (5863 images)
- Classes: Normal & Pneumonia
- Split: Train / Test / Validation
2. Preprocessing #
- Rescaling: 1./255
- Augmentation: Shear, Zoom, Flip
3. Model Building #
- Custom CNN: Conv2D → MaxPooling → Flatten → Dense
- Pretrained: VGG16, ResNet-50, AlexNet
- Activations: ReLU, Softmax/Sigmoid
4. Training #
- Platform: Google Colab
- Epochs: 25, Batch Size: 32
- Optimizer: Adam, Loss: Binary Crossentropy
5. Evaluation #
- Accuracy: 95%, Precision: 94.41%
- Metrics: Confusion matrix, F1 score
6. Deployment #
- Web Framework: Flask
- Live on render (link is down here)
- Upload X-ray → Predict → Show result
- Result: Pneumonia Positive/Negative + Confidence
Future Enhancements #
- Add clinical data (age, symptoms)
- Multi-disease detection (e.g. TB)
- Mobile app or cloud deployment
- Use segmentation to highlight infected areas
- Collaborate with medical institutions for real validation
#
Thank You! #
We sincerely appreciate your time and attention.
Questions & Discussion Welcome.



Loading discussion...