A Multi-Task Based Deep Learning Framework With Landmark Detection for MRI Couinaud Segmentation.

Area of Science:

• Medical Imaging
• Surgical Planning
• Computational Anatomy

Background:

• Precise Couinaud liver segmentation is crucial for hepatic surgery planning.
• Anatomical variations and complex liver structures pose challenges to accurate segmentation.
• Current methods may lack the robustness needed for diverse patient populations and imaging conditions.

Purpose of the Study:

• To develop and validate a novel automated approach for Couinaud liver segmentation using CE-MRI.
• To enhance the precision of preoperative planning for hepatic surgery.
• To improve the adaptability of segmentation to anatomical variability and reduce postoperative complications.

Main Methods:

• Utilized a multi-task learning framework for synchronized landmark detection and liver segmentation.
• Employed portal venous phase contrast-enhanced magnetic resonance imaging (CE-MRI) data.
• Identified seven key anatomical landmarks to guide segmentation.

Main Results:

• Achieved an average Dice Similarity Coefficient (DSC) of 85.29% for Couinaud segment delineation.
• Outperformed existing models by 3.12% in segmentation accuracy.
• Demonstrated robust performance across diverse patient groups (normal, diseased livers) and varied imaging parameters (field strengths, devices, contrast agents).

Conclusions:

• The developed automated method offers a pioneering solution for Couinaud liver segmentation using CE-MRI.
• Correlating landmark detection with segmentation enhances surgical planning accuracy and robustness.
• This technique holds significant potential for improving clinical outcomes in hepatic surgery by enabling tailored interventions and minimizing risks.