AI-based mobile applications for public health response

details

With today’s increased adoption of mobile smart devices by citizens, there is a clear trend for them to use such devices to deal with their more critical data. Currently, personal,financial and even health information is collected, processed and stored in these smart devices. In particular, the variety of sensors present on either smartphone, smart watches or smart bands, some of them collecting physiological data, has increased substantially, and as a consequence, user health data collection has reached a level never seen before – the common gateway that aggregates all this data is the user smartphone (paired with other end-user devices).

In a time where the SARS COV-2 virus (COVID-19) spreads around the world, responsible authorities as well as stakeholders of the public and private health sectors, are turning to the development of solutions that might help prevent the further spread of the virus. One of current technological trends, refers to the applicability of contact tracing applications on the user’s smartphones, to inform users and health authorities about potential or actual spread. This approach raises user’s security and privacy concerns, including protection of personal data, particularly, in face of privacy legislation and regulation, particularly the GDPR, enforced by the European Union since 25 May 2018.

To contribute with a preventive approach for Public Health strategies in facing such pandemic situation, while encompassing such concerns, our project proposes a smartphone app and trustworthy Artificial Intelligence (AI) distributed and service-based platform, to identify symptomatic and asymptomatic patients as well as assess the risk of exposure, where smartphone collected data, including health data, will be securely stored on the device (using a secure data vault), and the access to such secure data vault will need to be controlled and authorized by the user in his/her own device.

The basic trust principles are threefold: user-centric, transparent, and adherence to open-source.We envisage addressing 2 scenarios:

(1) risk Groups, i.e. diabetic, hypertensive patients, with heart disease. In this scenario, we plan to monitor various discontinuous physiological variables (coughing, breathing, temperature, heart rate, O2 saturation, physical activity);
(2) COVID-19 patients in home care, with the need for supplementary hardware to register the same physiological variables, possibly continuous, 24/7.

For both scenarios we will collect also social media data. To serve these scenarios, we will develop a smartphone app and an AI secure back end adopting a federated edge computing architecture based on micro-services. With such system, we will collect data in a secure way, which will be analyze dusing AI methodologies (particularly, Machine Learning, ML), aiming at aiding and alerting the user and his/her doctor regarding clinical situations such as:

(1) establishing a diagnosis of COVID-19;
(2) assessing the risk of being infected by the virus and advising the user, as well as other citizens (in a fully anonymized way), with whom he/she has been recently in contact, to perform a COVID-19 test.

In this user-centric privacy ecosystem, the smartphone will act as a generic health data collection, aggregation and storage device, and can be paired with external devices. The smartphone will store all the user’s sensitive health data on a securely encrypted data vault (using multi-factor unlocking mechanisms) and will act as a personal generic data gateway between the data processors (particularly, large public hospitals of the National Health System) and the user anonymized data. The smartphone app will use AI and distributed ledger technologies based on blockchain technology, which will act as common data repository, recording all the different processing performed to the user data by a set of entities, able to build trustworthy ways to detect and interpret personal health, avoiding unwanted communication as much as possible.

Due to its public nature, no private data will be stored on the blockchain, only specific data about the conducted operation and by whom. Also, through the usage of smart contracts, we will bind the permissions given by the users to data processors, to conduct certain operations over their health data. Our AI approach, will help to sort through the data to get reliable and accurate insights about changes in usual rhythms that indicate the need of medical intervention, acting as a remote health monitoring system.

This project envisions the consideration of federated ML approaches for ongoing, incremental training, as well as agile inference (classification) mechanisms for fast, private and secure identification of symptomatic and asymptomatic patients as well as assessment the risk of exposure. We will use a large scale public data center available at RNCA (Rede Nacional de Computacão Avançada), during development, including for hyper-parameter search and network tuning.