PERFORMANCE EVALUATION OF ASYMMETRIC MULTICORE PROCESSORS USING ARM BIG.LITTLE ARCHITECTURE IN MRI IMAGING RECONSTRUCTION

Abstract

MRI image reconstruction is a crucial process in medical diagnosis that requires high accuracy and computational efficiency. This research uses ARM big.LITTLE architecture on Orange Pi 5 device to improve the efficiency of MRI image reconstruction with Inverse Fast Fourier Transform (IFFT) algorithm. The k-space data of 512x512, 1024x1024, and 2048x2048 are tested in serial (big_CPU and LITTLE_CPU) and parallel (big_CPU, LITTLE_CPU with 4 threads, and hybrid bigLITTLE and LITTLEbig up to 8 threads) scenarios. Results show that big_CPU with 4 threads is optimal for 512x512 k-space, bigLITTLE with 8 threads is best for 1024x1024 k-space, and LITTLEbig with 8 threads is most efficient for 2048x2048 k-space, with speed-ups reaching 3,93 times. The hybrid configuration proved superior for medium to large data processing, while big_CPU remains ideal for small data. This research confirms the potential of the ARM-based AMP architecture.LITTLE in optimizing MRI image reconstruction, providing a balance between high performance and energy efficiency for medical applications.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint