Lung Localization in Chest X-ray Images

This repository contains the implementation for the Digital Image Processing (PCD) course final project, titled "Lung Area Localization in Chest X-ray Images Using Progressive Distance-Based Masking and Anatomical Patterns".

Paper

Read the full paper here:

https://doi.org/10.5281/zenodo.18448574

Overview

This project implements a classical computer vision approach to segment lung areas in Chest X-Ray (CXR) images. Unlike modern Deep Learning approaches (U-Net, etc.), this method relies on anatomical priors, morphological operations, and progressive distance-based masking to achieve robust segmentation without heavy training data.

Key Features

• Preprocessing: Adaptive histogram_equalization (CLAHE) and Gaussian smoothing.
• Body Extraction: Thresholding and morphological reconstruction to isolate the body area.
• Segmentation: Novel distance-based thresholding and watershed algorithm to separate lungs.
• Analysis: Rib pattern detection using autocorrelation and FFT to validate lung regions.

Project Structure

project/
├── data/               # Input Chest X-Ray images
├── notebooks/          # Jupyter notebooks for experiments and visualization
├── results/            # Output segmentation masks and evaluation metrics
├── scripts/            # Helper scripts (demo, comparison, data organization)
├── src/                # Core source code package
│   ├── analysis/       # Rib detection, periodicity, and pattern analysis
│   ├── preprocessing/  # Image enhancement (CLAHE) and body mask extraction
│   ├── segmentation/   # Distance-based thresholding and watershed logic
│   └── utils/          # Visualization tools and image processing helpers
├── tests/              # Unit tests
├── main.py             # Main entry point for running the pipeline
├── requirements.txt    # Python dependencies
└── setup.py            # Package installation script

Folder Descriptions

• src/preprocessing/: Contains body_extraction.py and enhancement.py. Handles the initial image processing chain including smoothing, CLAHE, and generating the initial coarse body mask.
• src/segmentation/: Core logic for the progressive distance-based masking (distance_based.py) and watershed algorithms (watershed.py) to refine lung boundaries.
• src/analysis/: Includes rib_detection.py and periodicity.py to analyze texture patterns within the segmented regions, ensuring they match anatomical expectations of ribs.
• src/utils/: General utilities for visualization.py (plotting results) and image_utils.py (array manipulations).
• scripts/: executable scripts for specific tasks like running demos (demo.py), comparing methods (compare.py), or organizing datasets.
• notebooks/: Interactive environments for testing individual functions and visualizing intermediate steps.

Results

*Figure 1: Performance visualization on the JSRT dataset. The image displays a comparison of the 5 Best (top row), Average (middle row), and Worst (bottom row) results.

Installation

1. Clone the repository:

   git clone https://github.com/fae-ld/Lung-Localization
   cd Lung-Localization

2. Install dependencies:

   pip install -r requirements.txt

   Requirements:

   • numpy
   • matplotlib
   • scikit-image
   • scipy
   • opencv-python

Usage

You can run the main pipeline using the provided main.py script:

python main.py

Or run the specific demo script:

python scripts/demo.py

Example Code

To use the segmentation pipeline in your own script:

from src.preprocessing.body_extraction import extract_body_mask
from src.segmentation.distance_based import apply_distance_thresholding, calculate_8_way_distances
from skimage import io

# Load image
img = io.imread("data/sample_cxr.png", as_gray=True)

# 1. Extract Body
body_mask, _, _ = extract_body_mask(img)

# 2. Calculate Distance Map
dist_map = calculate_8_way_distances(body_mask)

# 3. Apply Segmentation
lung_mask = apply_distance_thresholding(dist_map, method='sigma')

License

MIT License

Acknowledgments

Note: This project is mainly exploratory and is open to discussion and collaboration.