Detecting Congenital Heart Defects on an Edge platform using Phonocardiogram and AI

Congenital Heart Defects (CHD) are heart malformations caused by abnormal heart development. These malformations can induce a wide range of clinical symptoms, indicating that this vital organ is underperforming. The earlier one can detect these malformations, the better patient outcomes. Ultrasound is used for the early detection of CHD. However, because antenatal and postnatal ultrasounds are in short supply, diagnosis is generally performed based on other solutions which are more easily available. In a resource-constraint context, where ultrasound screening is highly limited, these alternative methods may become even more crucial. Routine CHD screening is executed in such settings by the mean of a multi-dimensional clinical test that includes, among others, pulse oximetry and auscultation. Although auscultation is subject to interpretation, as some cardiac aberrations are not always audible, it facilitates cardiac defects detection. Some novel Artificial Intelligence (AI) driven methodology for the detection of CHD has been developed at the Embedded.Systems@UCC research group. The purpose of this study is to implement a clinical decision-making device relying on AI to aid in the clinical differentiation of sounds affected by CHDs. This work includes implementation and evaluation for Machine Learning (ML) based Segmentation, Feature Extraction and Classification on a Raspberry Pi device. In fact, validation of these techniques on EDGE IoT (Internet of Things) devices is paramount towards the early detection of CHDs. The final goal is to realise a first Demo of a portable, rapid, and low-cost Phonocardiogram (PCG) signals real-time monitoring system that merges the previous research into a unique device. The equipment chosen for this purpose is: - Raspberry Pi model 4 (RPi4): a single board computer that saves the received soundtracks through a jack port and processes them with a Python algorithm. - Thinklabs One Digital Stethoscope: this medical device records and shares sounds, providing various solutions for Telemedicine, Education, Research and Electronic Medical Records (EMR). It is important to underline that this implementation serves as the initial demonstration of the entire research project. Indeed, the developed system is optimised only in terms of the execution time of each section. The basic idea and demo will support additional optimisation for the algorithm and the embedded system.