Time-Optimized Detection of Cardiovascular Complications with Artificial Intelligence in Rescue Operations using FPGA-based Wearable

Recent research on Artificial Neural Networks (ANNs) has shown significant improvement in machine learning over traditional algorithms in many disciplines. This paper contributes to the advances in medical science and AI technologies by exploring this promising technology for real-time cardiovascular complication detection and resuscitation during rescue missions. Previous studies have relied on cloud-based computing or specialized hardware such as graphics processing units (GPUs), which can be expensive and require significant power consumption. Additionally, existing AI models are often not optimized for low-latency processing, hindering their real-time applications. This study proposes a PyTorch-based ANN model with time optimization techniques on the field-programmable gate arrays (FPGAs) hardware platform, providing data privacy and hardware security. Our approach includes intermediate layer saving and layer parameter reuse, reducing computational complexity and memory requirements while maintaining accuracy. The prototype wearable utilizes a Trenz Electronics TE0802 FPGA board with custom PYNQ-Linux software, providing a low-cost, low-power, and high-performance hardware platform. Using the Apache TVM toolchain, our ANN model predicts cardiovascular disease risk and aids rescuers in making rapid and precise clinical decisions. The results demonstrate 95.9% accuracy in detecting cardiovascular complications, with an average execution time of 41.99ms using TVM. Additionally, our time optimization technique achieves reduced inference times of 33%, 55%, and 79% for reusing the saved output files of layers 1, 2, and 3, respectively, as validated through simulations and experiments.