H-SegNet: hybrid segmentation network for lung segmentation in chest radiographs using mask region-based...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Radiology

Background:

• Chest x-ray (CXR) is a primary tool for pulmonary disease detection.
• Accurate lung segmentation is essential for computer-aided diagnosis systems using digital CXRs.
• Challenges in lung segmentation include low contrast and inter-subject variability.

Purpose of the Study:

• To develop an automatic Hybrid Segmentation Network (H-SegNet) for precise lung segmentation in CXRs.
• To address the difficulties posed by low contrast and anatomical variations in lung boundary detection.
• To enhance the accuracy of computer-aided detection and diagnosis systems through improved lung segmentation.

Main Methods:

• Proposed a two-step approach: deep learning-based preprocessing for coarse lung contour extraction.
• Implemented a refinement step combining an improved principal curve method and a machine learning technique.
• Developed an automatic Hybrid Segmentation Network (H-SegNet) for end-to-end lung segmentation.

Main Results:

• H-SegNet demonstrated superior performance in segmenting lung regions on CXR images.
• The method achieved state-of-the-art segmentation results across multiple public datasets.
• Experimental validation confirmed the effectiveness of the proposed deep learning and refinement strategy.

Conclusions:

• The H-SegNet offers a robust and accurate solution for automatic lung segmentation in chest x-rays.
• This advancement can significantly improve the reliability of computer-aided diagnostic tools for pulmonary conditions.
• The hybrid approach effectively overcomes common challenges in CXR-based lung segmentation.