Lung Field Segmentation in Chest Radiographs From Boundary Maps by a Structured Edge Detector

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Accurate lung field segmentation is crucial for automated analysis of chest radiographs (CXRs).
• Existing methods may face challenges in efficiency and accuracy.
• Developing robust segmentation techniques is essential for advancing diagnostic capabilities.

Purpose of the Study:

• To present a novel and efficient method for lung field segmentation in CXRs.
• To leverage structured edge detection (SED) for high-quality boundary mapping.
• To evaluate the accuracy, speed, and generalizability of the proposed segmentation technique.

Main Methods:

• A structured edge detector (SED) was trained to identify lung boundaries in CXRs.
• An ultrametric contour map (UCM) was generated from the SED's boundary map.
• Lung contours were extracted from the UCM based on confidence levels.

Main Results:

• The method achieved an average Jaccard index of 95.2% on the JSRT database.
• Segmentation computation time was under 0.1 seconds per CXR on a standard laptop.
• Validation on diverse digital radiography units demonstrated model generalization.

Conclusions:

• The proposed SED-based method offers a highly accurate and computationally efficient solution for lung field segmentation in CXRs.
• The technique shows strong generalization capabilities across different imaging equipment.
• This approach holds significant potential for improving automated medical image analysis workflows.