Semantic Segmentation of CT Liver Structures: A Systematic Review of Recent Trends and Bibliometric Analysis : Neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• AI for medical image segmentation is a growing research area.
• Previous studies show varied AI tool performance in liver segmentation.
• A systematic overview of this field is lacking.

Purpose of the Study:

• To systematically review and bibliometrically analyze AI, particularly deep learning, for liver structure segmentation in CT images.
• To outline research directions based on algorithmic features.
• To assess algorithm modeling objectives, performance benchmarks, and model complexity.

Main Methods:

• Systematic and bibliometric review of relevant publications.
• Focus on fully automatic semantic segmentation of liver structures in CT scans.
• Analysis of algorithmic features, performance, and complexity.

Main Results:

• Fully automatic hybrid 2D and 3D networks are top performers for general liver segmentation.
• Fully automatic generative approaches show the best performance for liver tumor and vasculature segmentation.
• Current benchmarks indicate room for improvement in segmenting small structures in high-resolution CT scans.

Conclusions:

• Hybrid 2D/3D networks are leading in liver segmentation.
• Generative AI is promising for complex liver structures like tumors and vasculature.
• Further research is needed to enhance AI accuracy for small, detailed structures in abdominal CT imaging.