Abstract:
This thesis investigates the utilization of FPGA technology in tuberculosis (TB) detection through X-ray image analysis, focusing on designing an efficient system within the Kria KV260 FPGAbased System-on-Chip architecture. The methodology integrates hardware development emphasizing the Kria KV260 FPGA-based SoC and deep learning model training, transitioning from conventional computing to specific optimization within Vitis AI for deployment on the Kria KV260 platform. Comparative analysis assesses the FPGA-based SoC against CPU-based platforms, evaluating power efficiency, throughput, and latency. Results highlight the FPGA's superiority, notably the Kria KV260, demonstrating significantly lower power consumption during inference, emphasizing FPGAs inherent energy efficiency. Additionally, the Kria KV260 exhibits superior throughput and lower latency, crucial for efficient TB detection, underscoring FPGAs speed and responsiveness. In conclusion, the study extensively demonstrates FPGA-based architectures' prowess, particularly the Kria KV260, in enhancing energy efficiency and performance for TB detection from X-ray images. These findings emphasize substantial improvements in power consumption, throughput, and latency compared to conventional CPUbased platforms, particularly vital in healthcare applications reliant on deep learning inference. This research lays a foundation for future advancements leveraging FPGA technology for efficient and accurate TB diagnosis in healthcare settings.
