FPGA-Based Medical Image Processing Using Hardware-Software Co-Design Approach.

Area of Science:

• Biomedical Engineering
• Computer Engineering
• Medical Imaging

Background:

• Medical image analysis for diseases like Malaria and Pneumonia often requires significant computational resources.
• Real-time processing is crucial for timely diagnosis and treatment in clinical settings.
• Existing hardware solutions may face limitations in performance and energy efficiency.

Purpose of the Study:

• To develop and evaluate a Field-Programmable Gate Array (FPGA) based framework for accelerated medical image processing.
• To implement a hardware-software co-design approach for efficient Malaria and Pneumonia detection.
• To optimize the architecture for high throughput and low power consumption.

Main Methods:

• Utilized an AMD-Xilinx UltraScale+ MPSoC (ZCU104) FPGA for implementation.
• Employed a customized high-level synthesis (HLS) process to optimize data movement between Processing System (PS) and Programmable Logic (PL).
• Incorporated depth-wise convolution, layer fusion, and custom cache for computational and memory access efficiency.

Main Results:

• Achieved a throughput of 298.22 FPS for Malaria detection and 205.87 FPS for Pneumonia detection.
• Demonstrated significant energy efficiency with 14.62 mJ/img for Malaria and 23.89 mJ/img for Pneumonia.
• Outperformed Raspberry Pi with Coral TPU by 8.3× in throughput and 4.3× in energy efficiency.

Conclusions:

• The proposed FPGA-based framework offers a powerful and energy-efficient solution for real-time medical image analysis.
• Hardware-software co-design and architectural optimizations are effective in accelerating biomedical tasks.
• This approach is highly suitable for deployment in resource-constrained or real-time medical diagnostic systems.