Embedded DNN Classifier for Five Different Cardiac Diseases

The evolution of modern healthcare has been significantly shaped by the convergence of connected sensors, smart Wearable Devices, Artificial Intelligence, and the Internet of Things giving rise to the domain of eHealth and offering invaluable insights into the complications of heart health. eHealth's impact extends to facilitating diagnosis, treatment, and medication for a diverse array of conditions, prominently including cardiac diseases. Despite substantial strides in medical technology, the detection of arrhythmia remains a persistent challenge, with early diagnosis holding the potential to avert numerous fatalities. This paper proposes an ultra-lightweight (876KB) Embedded-Deep Neural Network model specifically designed for resource-constrained devices. With high accuracy ranging from 94% to 99% for five classes identified from the MIT-BIH dataset, the proposed model is small enough to fit on tiny devices like the Arduino Nano BLE 33 Sense. This translates to low power consumption and real-time inference, making it ideal for screening cardiac diseases on wearable devices.