Comprehensive Review on the Exploitation of Advanced Memory Optimization Strategies to Improve Performance for Convolutional and Spiking Neural Networks in Medical Imaging Using Hardware Accelerators

Abstract

Advanced memory optimization techniques are reviewed to enhance the performance of Convolutional Neural Networks (CNNs) and Spiking Neural Networks (SNNs) on hardware accelerators, addressing the real-world challenges in medical imaging. This review evaluates various platforms: In-Memory Computing (IMC), Field-Programmable Gate Array (FPGA), Python Productivity for Zynq (PYNQ-Z2), Graphics Processing Unit (GPU), and Application-Specific Integrated Circuit (ASIC) concerning overcoming memory bottlenecks, minimizing latency, and reducing energy consumption in Magnetic Resonance Imaging (MRI) reconstruction, Computed Tomography (CT) scan analysis, and real-time diagnostics. It will analyze techniques like memory compression, tiling, hierarchical memory management, and neural network pruning to improve computation efficiency. In addition, in-memory computing will be a key focus to mitigate the inefficiency of data movement, adaptability of Field-Programmable Gate Array (FPGA) for custom workloads, parallel processing by Graphics Processing Unit (GPU), and domain-specific optimizations of Application-Specific Integrated Circuit (ASIC). This review addresses the challenges of high-resolution image processing and energy constraints to provide a comprehensive guide to scalable, efficient hardware accelerators for neural networks in medical imaging.

Keywords

• Memory optimization
• neural networks
• FPGA
• in-memory computing
• medical imaging
• hardware accelerators