GGO nodule volume-preserving nonrigid lung registration using GLCM texture analysis

Area of Science:

• Medical imaging analysis
• Radiology
• Computer-aided diagnosis

Background:

• Accurate lung nodule assessment in cancer screening relies on comparing follow-up CT scans.
• Nonrigid registration is crucial for aligning scans, but preserving nodule volumes, especially for ground glass opacity (GGO) nodules, is challenging due to their inhomogeneity.
• Existing methods struggle with automatic segmentation and volume preservation of GGO nodules.

Purpose of the Study:

• To develop an accurate and fast nonrigid registration method for lung cancer screening.
• To specifically address the challenge of volume preservation for ground glass opacity (GGO) nodules.
• To accelerate the registration process using GPU computing.

Main Methods:

• Proposed a nonrigid registration method applying a volume-preserving constraint to candidate GGO nodule regions.
• Utilized gray-level co-occurrence matrix (GLCM) texture analysis and intensity values for automatic detection of GGO nodule candidates.
• Accelerated computationally intensive registration steps (image transformation, cost-function calculation) using Compute Unified Device Architecture (CUDA).

Main Results:

• The method achieved near-perfect volume preservation of GGO nodules in the floating image.
• Effectively aligned the lungs between reference and floating computed tomography images.
• Demonstrated a significant acceleration, achieving approximately 20x faster registration compared to conventional methods using CUDA.

Conclusions:

• The developed method accurately and efficiently performs volume-preserving nonrigid registration for GGO nodules in lung cancer screening.
• The approach shows promise for improving nodule follow-up studies and can be extended to other organs and diseases, such as liver cancer.
• CUDA acceleration significantly enhances the computational performance of the registration process.