AI-Powered Clinical X-Ray Diagnostic System

An end-to-end deep learning pipeline for multi-label classification of 14 chest pathologies from frontal chest X-ray images.
This project evolves from raw data training to a calibrated, explainable, and automated clinical reporting system.

---

Project Highlights

Architecture

• Fine-tuned DenseNet-121 backbone
• Global Average Pooling head
• Multi-label sigmoid output for 14 pathologies

Explainability (XAI)

• Integrated Grad-CAM visualization
• Heatmap overlays to verify model attention regions
• Clinically interpretable outputs

Clinical Reliability

• Probability calibration applied post-training
• Reduced overconfidence in predictions
• Improved Expected Calibration Error (ECE)

Automation

• Batch inference engine
• Structured CSV clinical report generation
• Scalable processing for hundreds of X-ray images

---

Performance Metrics

The model demonstrates expert-level discrimination ability across several critical pathologies.

Pathology	AUC Score
Cardiomegaly	0.926
Emphysema	0.922
Pneumothorax	0.901
Effusion	0.897
Average AUC	0.844

---

Reliability Calibration

Post-training probability calibration significantly reduced the Expected Calibration Error (ECE).

• Near 0.00% ECE across most classes
• Improved probability reliability
• Reduced model overconfidence
• Better alignment with real clinical prevalence

---

Visual Evidence (Grad-CAM)

The system not only predicts pathologies but also provides visual explanations.

Workflow:

1. Input: Standard frontal chest X-ray
2. Model prediction
3. Grad-CAM heatmap generation
4. Overlay visualization highlighting suspicious regions

Output:

• Red/Yellow heatmaps indicate regions responsible for positive predictions
• Supports clinical interpretability and model trust

---

Key Features

• Automated Reporting

  • Processes entire image directories
  • Exports final_clinical_report.csv

• Statistical Dashboard

  • Pathology distribution charts
  • Frequency bar plots and pie charts

• Real-time Progress Tracking

  • Integrated tqdm for batch inference

• XAI Visualizer

  • Automatic Grad-CAM heatmap generation
  • Saved per detected abnormality

---

System Pipeline

Raw X-ray Images
        ↓
Preprocessing & Normalization
        ↓
DenseNet-121 Multi-Label Classifier
        ↓
Probability Calibration
        ↓
Grad-CAM Explainability
        ↓
Batch Inference Engine
        ↓
CSV Clinical Report + Visual Outputs

---

Tech Stack

• Python
• PyTorch
• Torchvision
• NumPy / Pandas
• Matplotlib / Seaborn
• OpenCV
• tqdm

---

Clinical Disclaimer

This project is a research-grade AI system intended for academic and experimental purposes only.
It is not approved for clinical deployment and must not be used as a substitute for professional medical diagnosis.