Accelerating CNNs for Pneumonia Disease Diagnosis via Heterogeneous FPGA Systems

Early and accurate diagnosis of pneumonia is critical for effective treatment and improved patient outcomes. This paper introduces a cutting-edge framework for accelerating the diagnosis of pneumonia using Convolutional Neural Networks (CNNs) on heterogeneous Field-Programmable Gate Array (FPGA) systems. The framework is designed to leverage FPGAs’ parallel processing capabilities and reconfigurability, enabling significant enhancements in speed and efficiency for processing medical images. Our approach involves the development of an optimized CNN model tailored to the features and characteristics typical of pneumonia in radiographic images. This model is then mapped onto an FPGA, utilizing high-level synthesis and custom hardware descriptions to maximize data throughput and minimize latency. We conducted testing using a diverse chest X-ray dataset, comparing our FPGA-based system’s performance against traditional computing platforms such as GPUs and CPUs. Results indicate that our system accelerates the diagnostic process over 5 times and achieves a 97.58% accuracy rate comparable to state-of-the-art methods. The proposed framework enhances medical imaging devices’ capabilities, leading to quicker diagnoses, thus reducing hospitalization time and improving patient care.