🧠 Multi-Model Classification for Chest X-Ray Images

This repository provides a modular and extensible training pipeline for comparing multiple deep learning models on a chest X-ray dataset with four distinct classes. The primary goal is to evaluate different CNN architectures in terms of accuracy, training efficiency, and diagnostic performance using various classification metrics.

📌 Features

• ✅ Supports 27+ CNN architectures from torchvision.models:
  • ResNet (18, 34, 50, 101, 152)
  • DenseNet (121, 161, 169, 201)
  • VGG (11, 13, 16, 19 with/without batch norm)
  • MobileNetV2, MobileNetV3 (Large/Small)
  • EfficientNet (B0, B1)
  • AlexNet, SqueezeNet (1.0, 1.1)
  • ShuffleNetV2 (x0.5, x1.0)
  • GoogLeNet, InceptionV3 (with auxiliary heads)
• 🏥 Classifies chest X-ray images into four categories (e.g., Normal, Pneumonia, COVID-19, Tuberculosis)
• 📊 Tracks key evaluation metrics:
  • Accuracy
  • Precision
  • Recall
  • F1 Score
  • Confusion Matrix
• 💾 Saves:
  • Model weights (.pt)
  • Training history as CSV
  • Confusion matrix images
  • Final comparison summary across models
• ⚙️ Handles special architectures like Inception and GoogLeNet (auxiliary outputs)
• 📈 Automatically generates plots of training/validation metrics

📂 Dataset

The dataset used for this project can be found on Kaggle:

🔗 Chest X-ray Pneumonia, COVID-19, Tuberculosis Dataset – Kaggle

Folder Structure

Before training, structure the dataset as follows:

data/
├── train/
│ ├── Normal/
│ ├── Pneumonia/
│ ├── COVID-19/
│ └── Tuberculosis/
├── val/
│ ├── Normal/
│ └── ...
└── test/
├── Normal/
└── ...

Each subdirectory should contain images belonging to that class. This layout is compatible with PyTorch's ImageFolder.

---

📊 Outputs

After training, the following files will be saved inside the results/ directory:

results/
├── resnet18_history.csv               # Training & validation metrics
├── resnet18_confusion_matrix.png     # Confusion matrix image
├── resnet18.pt                        # Trained model weights
├── ...
summary.csv                            # Final metric comparison across all models

---

🧪 Evaluation Metrics

All models are evaluated using:

• Accuracy
• Precision
• Recall
• F1 Score
• Confusion Matrix

---

🧾 Citation / Attribution

If you use this repository, code, or any part of it in your research, publication, or project, please cite it or provide appropriate credit:

Emirkan Beyaz, "Multi-Model Evaluator", GitHub Repository, 2025.
https://github.com/Hords01/multi-model-evaluator

✉️ Contact

For questions, issues, or collaboration opportunities, feel free to open an Issue or start a Discussion on GitHub. mail:beyazemirkan@gmail.com