[Hippocampus MRI Parallel Segmentation Using Three Dimensions Lattice Boltzmann Model with Prior Information].

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Artificial Intelligence

Background:

• Accurate hippocampus volume measurement is crucial for diagnosing Alzheimer's disease (AD) and other neurological conditions.
• Traditional 3D segmentation methods can be computationally intensive, limiting their clinical application.
• Leveraging spatial and intensity correlations in MRI data enhances segmentation accuracy.

Purpose of the Study:

• To develop a novel, computationally efficient 3D segmentation method for hippocampus volume analysis.
• To improve the accuracy and speed of brain MRI segmentation for neurodegenerative disease diagnosis.
• To validate the performance of a new Lattice Boltzmann (LB) model combined with deformable models and prior information.

Main Methods:

• A novel 3D Lattice Boltzmann (LB) model was developed, integrating surface evolution of deformable models.
• Prior information was incorporated as an external force term to constrain 3D surface evolution.
• The method was parallelized on single and dual Graphics Processing Unit (GPU) platforms to address computational costs.

Main Results:

• The novel LB method achieved high segmentation accuracy for hippocampus volume.
• Significant reductions in computational time were observed: 12.76s (single GPU) and 17.32s (dual GPU), compared to 132.43s (CPU).
• The parallelized LB method demonstrated high computational efficiency and accuracy on real AD patient MRI data.

Conclusions:

• The proposed 3D LB model offers an accurate and computationally efficient solution for hippocampus segmentation.
• GPU parallelization effectively addresses the computational challenges of 3D MRI segmentation.
• This method holds promise for accelerating the diagnosis and monitoring of Alzheimer's disease and related disorders.