Multi-Device Parallel MRI Reconstruction: Efficient Partitioning for Undersampled 5D Cardiac CINE.

Area of Science:

• Medical Imaging
• Computational Science
• Cardiovascular Diagnostics

Background:

• Cardiac cine MRI is vital for diagnosing heart conditions, but high-resolution imaging generates large datasets.
• Processing these large datasets poses computational challenges, potentially slowing down diagnostic imaging efficiency.
• Existing single-GPU systems face memory limitations when handling high-resolution, five-dimensional cardiac MRI data.

Purpose of the Study:

• To develop and evaluate a multi-GPU system for accelerated cardiac MRI reconstruction.
• To overcome single-GPU memory limitations for processing large, high-resolution cardiac MRI datasets.
• To enhance the efficiency and speed of cardiac MRI image reconstruction for improved diagnostic capabilities.

Main Methods:

• Utilized a multi-Graphics Processing Unit (GPU) system for parallel processing of cardiac MRI data.
• Implemented data partitioning to manage large datasets across multiple GPUs, overcoming single-device memory constraints.
• Employed OpenCL technology for cross-platform compatibility and system adaptability.

Main Results:

• The multi-GPU system significantly accelerated the reconstruction process for high-resolution, five-dimensional cardiac MRI.
• The approach successfully processed substantial data volumes while preserving image integrity.
• Demonstrated enhanced efficiency in generating cardiac MRI images compared to single-device limitations.

Conclusions:

• The proposed multi-device GPU approach effectively addresses computational challenges in cardiac MRI.
• This advancement accelerates image reconstruction, facilitating faster and more effective cardiac health assessments.
• The system offers a scalable and adaptable solution for modern medical imaging demands.