Lung Nodule Segmentation with a Region-Based Fast Marching Method

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Accurate lung nodule segmentation in computed tomography (CT) is crucial for lung cancer diagnosis.
• Challenges arise from nodule diversity and visual similarity to surrounding tissues.
• Robust segmentation is vital for effective computer-aided diagnosis (CADx) systems.

Purpose of the Study:

• To develop and evaluate a novel segmentation algorithm for lung nodules in CT data.
• To assess the algorithm's accuracy across different nodule types (solid, non-solid, cavitary, round, irregular).
• To compare the proposed method against existing techniques, including active contour models and deep learning networks.

Main Methods:

• A segmentation algorithm combining the fast marching method with region growing and k-means clustering was developed.
• The algorithm segments images into similar feature regions, then merges them.
• Evaluation involved objective (Dice scores) and subjective methods on simulated and real patient data.

Main Results:

• The proposed method achieved high mean Dice scores for solid nodules (0.933 for round, 0.901 for irregular).
• Performance decreased for non-solid (0.799) and cavitary (0.614) nodules.
• The algorithm outperformed active contour models and showed comparable results to DBResNet, though less accurate than 3D-UNet.

Conclusions:

• The proposed fast marching-based segmentation method shows significant potential for lung nodule analysis in CADx.
• The technique is particularly effective for solid nodules, offering a promising tool for radiologists.
• Further development may be needed to improve segmentation accuracy for non-solid and cavitary nodules.