Segmentation of brain image volumes using the data list management library

Area of Science:

• Neuroimaging
• Computer Science
• Medical Image Analysis

Background:

• Accurate brain image segmentation is crucial for detecting neuroanatomical structures and monitoring brain lesions.
• High-resolution brain imaging requires processing numerous 2D slices, demanding significant computational resources.
• Parallel processing is essential to manage the computational load of large-scale neuroimaging datasets.

Purpose of the Study:

• To present a novel approach for parallel segmentation of brain image volumes.
• To introduce and evaluate a new version of the Data List Management Library (DLML) for neuroimaging applications.
• To compare the performance of the DLML approach against traditional parallel methods like MPI.

Main Methods:

• Development of a new DLML version for parallel processing of image data.
• Organizing image identifiers into a list for parallel and dynamic load-balanced processing.
• Performance comparison using cluster configurations with 4-32 processors, evaluating DLML against MPI (master-slave and static data distribution).

Main Results:

• The DLML approach enables transparent parallel processing with dynamic load balancing.
• Demonstrated efficiency in handling large volumes of 2D brain slices for enhanced resolution.
• Comparative analysis showed performance metrics against MPI-based parallel strategies.

Conclusions:

• The DLML-based approach offers an efficient and scalable solution for parallel brain image segmentation.
• Dynamic load balancing in DLML simplifies parallel implementation for researchers.
• This method facilitates high-resolution neuroimaging analysis and lesion quantification.