A low-interaction automatic 3D liver segmentation method using computed tomography for selective internal radiation...

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate liver volume estimation is crucial for effective selective internal radiation therapy (SIRT).
• Traditional liver segmentation methods can be time-consuming and user-dependent.
• Minimizing human interaction in segmentation is key for clinical workflow efficiency.

Purpose of the Study:

• To introduce a novel, minimally interactive liver segmentation algorithm for precise anatomic liver volume estimation.
• To evaluate the accuracy, consistency, and user independence of the proposed segmentation approach.
• To facilitate improved treatment planning for selective internal radiation therapy (SIRT).

Main Methods:

• Integration of a localized contouring algorithm with a modified k-means method for slice segmentation.
• Development of novel initialization masks using intensity-based region growing and VOI corrections for single-slice initialization.
• Validation using 34 liver CT scans, comparing volumetric estimations against a manual gold standard.

Main Results:

• Achieved an average accuracy of 97.22% for volumetric calculations.
• Obtained an average Dice coefficient of 0.92, indicating high segmentation overlap.
• Demonstrated high consistency (P = 0.55) and independence from user initialization (P = 0.20, Fleiss' Kappa = 0.77 ± 0.06).

Conclusions:

• The novel liver segmentation algorithm provides accurate and consistent liver volume estimations.
• The method significantly reduces user dependency, enhancing its applicability in clinical settings.
• This automated approach holds promise for optimizing treatment planning in SIRT.