3D active surfaces for liver segmentation in multisequence MRI images

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Non-invasive diagnostic techniques like CT and MRI can replace biopsies.
• Accurate organ segmentation is crucial for surgical intervention planning.
• Automated liver segmentation, particularly from MRI, is challenging due to artifacts like noise and low contrast.

Purpose of the Study:

• To present a novel method for multichannel magnetic resonance imaging (MRI) automatic liver segmentation.
• To address the challenges of MRI liver segmentation, including artifacts and low contrast.

Main Methods:

• A 3D active surface minimization approach using a dual formulation of variational methods.
• Development of a compact descriptor integrating spatial and multisequence MRI information.
• Modeling the descriptor with a liver statistical model for volumetric regularization.

Main Results:

• The proposed method demonstrated fast and accurate 3D liver segmentation.
• Achieved a Dice Similarity Coefficient of 98.59% on 18 healthy liver studies.
• Outperformed state-of-the-art methods based on nine quality metrics.

Conclusions:

• The novel 3D active surface approach provides an accurate and efficient solution for automatic liver segmentation from multichannel MRI.
• The method's integration of a compact descriptor and statistical modeling improves segmentation quality and robustness.
• This technique holds potential for enhancing surgical planning and diagnostic accuracy in liver imaging.