Lung Nodule Detection under Artifact Conditions

🧠 Project Objective

This project leverages the pre-trained MONAI RetinaNet model for automatic lung nodule detection in CT scans, and evaluates how different image artifacts impact the model’s performance. It also quantitatively assesses image quality using standard metrics.

---

🧪 Artifacts Simulated

Three types of artifacts were introduced to simulate real-world challenges in medical imaging:

• Patient-based Artifact: Motion blur (simulating patient movement)
• Physics-based Artifact: Gaussian noise (simulating low-dose scans)
• Hardware-based Artifact: Ring artifact (simulating hardware-induced patterns)

---

📐 Evaluation Metrics

🔍 Image Quality

• PSNR (Peak Signal-to-Noise Ratio)
• CNR (Contrast-to-Noise Ratio)

🧬 Detection Accuracy

• Sensitivity (True Positives / Total Ground Truth Nodules)

---

⚙️ Pipeline Overview

1. Load CT series (DICOM)
2. Run MONAI RetinaNet detection on the clean volume
3. Add artifacts (motion, noise, ring) to simulate degraded scans
4. Re-run detection after degradation
5. Evaluate changes in:
   • Detection performance
   • Image quality metrics

---

📊 Example Results

The following figures demonstrate how detection is affected by each artifact:

✅ Original CT Slice with Ground Truth

🚶‍♂️ Motion Artifact + Detection

📉 Gaussian Noise + Detection

🎯 Ring Artifact + Detection

---

💡 Requirements

Install required dependencies:

 $ pip install monai torch pydicom numpy matplotlib pandas opencv-python scikit-image

🚀 How to Run

Run the script using:

 $ python main_script.py

📜 License and Attribution

This project includes components from the open-source MONAI Lung Nodule Detection Bundle, which is distributed under the Apache License 2.0.

We have retained the original README.md and license documentation inside the docs/ directory in compliance with MONAI's licensing terms.

Please credit MONAI if reusing or modifying this project.

Contributers