Digital remote health platform for MACE (Major Adverse Cardiac Event) prediction.
Developing a project whose core objective is to consume health tracking metrics (eg; Android Health Connect, Apple Health) data in order to power an early warning platform for heart attacks and adverse cardiac events generally. Achieved by applying ML algorithms and techniques evaluated in recent research, that only require vital signs easily measured by mobile devices along with user's demographic and health profile, to predict MACE probability in next 12 months with high accuracy [Alkhodari et al., 2020].
Anomalies past a probability threshold, require further medical triage and are alerted to
- the customer
- their nominated or subscribing network of family, friends and community members. Each communication is tailored to the preferred communication channel, language and health literacy of recipient, to ensure maximum engagement.
- remote or on-site care providers monitoring their customers from a dashboard
A loosely coupled architecture will encourage and allow affinity services to be integrated into the service.
Possible examples of future integration include:
- personal DNA related health information
- local emergency incidents impacting customer risk
Personalised AI assisted anomaly detection and AI assisted prediction at scale.
Harnesses the power of hyper-personalised AI, social networks, and mobile devices, wearables to remotely monitor at-home user's vital signs. With optional notification to user's nominated social network, providing them reassurance or escalating for action.
Leveraging an individualised AI model (the 'AI twin of user's health profile') that continuously learns about the user and what is 'normal' for them. All the time getting better at detecting anomalies by 'stacking' an AI model of user with the general model for range of users into an 'ensemble' model [Abdellatif et al., 2024]
Telehealth and remote monitoring services are driving down the required frequency of home care visits and hospital admissions and readmissions and associated costs for patient, informal caregivers and care providers. This service aims to complement the range of digital services that make this possible, in an as accessible way as possible, using user's personal device.
Future proofed
Component based architecture built round microservices will allow other complementary services to be built or bought 'off the shelf' and integrated into the application, to further refine the AI model's capabilities.
PoC technical overview
Please check out the readme in link below for more technical detail.
Infrastructure stack is currently AWS ECS, Kafka and Sagemaker.
Getting involved
Welcome collaboration and feedback, the Github project is at
https://github.com/jackpa99/iot_health_to_ai_aws
_References:
Alkhodari, M., Islayem, D.K., Alskafi, F.A. and Khandoker, A.H., 2020. Predicting hypertensive patients with higher risk of developing vascular events using heart rate variability and machine learning. IEEE Access, 8, pp.192727-192739.
Abdellatif, A., Mubarak, H., Abdellatef, H., Kanesan, J., Abdelltif, Y., Chow, C.O., Chuah, J.H., Gheni, H.M. and Kendall, G., 2024. Computational detection and interpretation of heart disease based on conditional variational auto-encoder and stacked ensemble-learning framework. Biomedical Signal Processing and Control, 88, p.105644._
Top comments (0)