This repository contains the graduation thesis project of Nguyễn Sinh Hùng from Ho Chi Minh Open University. The project focuses on building a deep learning-based screening system that classifies chest X-ray images to detect SARS-CoV-2 (COVID-19) infection and segment infected lung regions.
The COVID-19 pandemic has underscored the need for fast and accurate diagnostic tools. While RT-PCR remains the gold standard, its high cost and slow turnaround have led to interest in alternative approaches. Chest X-rays (CXR), being more accessible and affordable, are widely used but require expert interpretation.
This project proposes an AI-powered screening system that integrates multiple deep learning models:
- YOLOv8: Detects and localizes the lung regions in chest X-ray images.
- ResNet-50V2: Classifies the lung image into three categories: COVID-19, Non-COVID pneumonia, and Normal.
- U-Net: Segments infected regions in lungs if COVID-19 is detected.
- Develop a low-cost, fast, and accurate solution for COVID-19 screening using chest X-ray images.
- Provide an automatic classification pipeline to assist radiologists and clinicians.
- Enhance diagnostic insight by visualizing infected lung areas.
- COVID-QU-Ex: Chest X-ray dataset with lung segmentation masks.
- QaTa-COV19: High-quality CXR with annotated infected regions.
- COVIDx CXR-4: Large-scale open dataset (67,000+ CXR images).
The project follows a multi-stage AI pipeline:
| Model | Purpose | Input Size | Epochs | Metrics |
|---|---|---|---|---|
| YOLOv8n | Lung detection | 640×640 | 30 | mAP@50 = ~1.0 |
| ResNet-50V2 | Multi-class classification | 200×200 | 50 | Accuracy, F1, ConfMat |
| U-Net | Infection segmentation | 256×256 | 10 | IoU, Dice Score |
Hardware: Google Colab with NVIDIA A100 (40GB VRAM)
YOLOv8 achieved near-perfect detection accuracy (mAP@50 ≈ 1.0).
ResNet-50V2 fine-tuned model achieved high accuracy with balanced class performance.
U-Net successfully segmented infected lung regions with strong IoU and Dice metrics.
Client: React, Redux, TailwindCSS, Axios
Server: Flask
Database: Postgres, MinIO - simulate S3
- Add real-time analytics for epidemic monitoring
- Support mobile and cross-platform deployment
📽️ Demo Video: Drive Link
- 👨💻 Student: Nguyễn Sinh Hùng
- 🎓 Advisor: Dr. Lê Viết Tuấn
- 🏫 Faculty: Computer Science, Ho Chi Minh Open University
- 📅 Thesis Year: 2025
This project is developed for academic research. Please contact the author for any questions regarding data use or extension.