Efficient Edge AI for Neonatal Care: Implementing HIE Grading on Snapdragon Edge Device

Hypoxic-Ischemic Encephalopathy (HIE) requires prompt diagnosis within the first six hours post-encephalopathy injury to maximize treatment effectiveness. Efficient and accurate grading of HIE severity is therefore essential. Current clinical methods for HIE diagnosis are complex and time-intensive, demanding specialised skills and often require a multi-model diagnostic approach. In response to this challenge, we developed and implemented an automated HIE grading system to aid with clinical decision-making. The EEG signal is converted to the audio domain, with the spectrogram representation of the audio signal classified using an 8-bit integer quantized 2D Convolutional Neural Network (CNN). A high accuracy in differentiating mild and moderate abnormalities in the EEG signal. For integration into the clinical workflow and to enable real-time analysis, this system is implemented on the Qualcomm RB3 Gen 2 edge device. Robust performance suitable for practical application in neonatal intensive care settings is shown with an inference time of 62 ms. This approach advances HIE diagnosis by providing clinicians with an accessible and secure tool for precise and timely decision support which can run on their mobile device.