A Segmentation Method for Lung Parenchyma Image Sequences Based on Superpixels and a Self-Generating Neural Forest.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Image Segmentation

Background:

• Lung parenchyma segmentation is crucial for computer-aided diagnosis of lung nodules using CT.
• Existing methods struggle with complete segmentation, particularly in upper/lower lung regions and nodule-affected images, and exhibit slow processing speeds.

Purpose of the Study:

• To develop an improved lung parenchyma segmentation method for CT image sequences.
• To address limitations of existing methods in terms of completeness, speed, and accuracy, especially for complex cases.

Main Methods:

• Utilized lung parenchyma features to define Region of Interest (ROI) image sequences.
• Proposed a gradient and sequential linear iterative clustering algorithm (GSLIC) for ROI sequence segmentation and superpixel sample generation.
• Employed a genetically optimized Self-Guided Neural Field (SGNF) for superpixel clustering, using grey and geometric features for final segmentation.

Main Results:

• Achieved higher segmentation precision and accuracy compared to existing methods.
• Demonstrated an average processing time of 42.21 seconds per dataset.
• Obtained an average volume pixel overlap ratio of 92.22 ± 4.02% across four lung parenchyma image types.

Conclusions:

• The proposed method effectively segments all lung parenchyma image sequences with improved speed and accuracy.
• This advancement holds significant potential for enhancing the computer-aided diagnosis of lung nodules.