3-D segmentation of lung nodules using hybrid level sets

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Accurate lung nodule segmentation in CT images is crucial for malignancy assessment.
• Existing segmentation methods often struggle with segmenting large nodules effectively.

Purpose of the Study:

• To develop and validate a semi-automatic system for robust 3D lung nodule segmentation.
• To evaluate the system's performance on both small and large nodules with high accuracy.

Main Methods:

• A semi-automatic system employing anisotropic diffusion for denoising and a region of interest selection.
• Nodule segmentation using a Geodesic Active Contour model with a mean intensity threshold.
• Adaptive estimation of nodule mean intensity via image intensity histogram.

Main Results:

• Achieved a mean spatial overlap of 0.855 between segmented and reference nodules.
• Demonstrated a mean volume bias of 0.10 ± 0.2 ml and algorithm repeatability of 0.060 ml.
• Validated performance on diverse public databases, including nodules and phantoms.

Conclusions:

• The proposed method provides accurate and robust 3D segmentation for both small and large lung nodules.
• This system is suitable for reliable volume estimation, supporting clinical decision-making in lung nodule analysis.