Abstract
BackgroundTranslating research outputs into practical tools for medical practitioners is a neglected area and could have a substantial impact. One of the barriers to implementing artificial intelligence (AI) and machine learning (ML) applications is their practical deployment in the field. Traditional web-based (i.e., server sided) applications are dependent on reliable internet connections, which may not be readily available in rural areas. Native mobile apps require device specific programming skills as well as contemporary hardware and software, with often rapid and unpredictable platform specific changes. This is a major challenge for using AI/ML tools in resource-limited settings.MethodsAn emerging technology, progressive web applications (PWAs), first introduced by Google in 2015, offers an opportunity to overcome the challenges of deploying bespoke AI/ML systems. The same PWA code can be implemented across all desktop platforms, iOS and Android phones and tablets. In addition to platform independence, a PWA can be designed to be primarily offline.ResultsWe demonstrate how a neural network-based pneumonia mortality prediction triage tool was migrated from a typical academic framework (paper and web-based prototype) to a tool that can be used offline on any mobile phone—the most convenient deployment vehicle. After an initial online connection to download the software, the application runs entirely offline, reading data from cached memory, and running code via JavaScript. On mobile devices the application is installed as a native app, without the inconvenience of platform specific code through manufacturer code stores.DiscussionWe show that an ML application can be deployed as a platform independent offline PWA using a pneumonia-related child mortality prediction tool as an example. The aim of this tool was to assist clinical staff in triaging children for hospital admission, by predicting their risk of death. PWAs function seamlessly when their host devices lose internet connectivity, making them ideal for e-health apps that can help improve health and save lives in resource-limited settings in line with the UN Sustainable Development Goal 3 (SDG3).
Highlights
Research on the application of machine learning algorithms in the healthcare sector has grown quickly in recent years and aims to develop and improve disease diagnosis, prognosis, and treatment, including personalized medicine [1]. This in turn can accelerate the attainment of Sustainable Development Goal 3 (SDG3)
We aim to demonstrate a generic deployment process for neural networks using, as an example, a recent pneumonia mortality prediction machine learning algorithm [6] developed at the Medical Research Council Unit The Gambia at the London school of Hygiene and tropical Medicine (MRC Unit The Gambia at LSHTM)
The primary aim of this paper is to describe the deployment of a neural network model; a complete description of the machine learning algorithm is given in another publication [6]
Summary
Translating research outputs into practical tools for medical practitioners is a neglected area and could have a substantial impact. Research on the application of machine learning algorithms in the healthcare sector has grown quickly in recent years and aims to develop and improve disease diagnosis, prognosis, and treatment, including personalized medicine [1]. This in turn can accelerate the attainment of Sustainable Development Goal 3 (SDG3).. As the vast majority of childhood Pneumonia deaths occur in low and middle income countries [7], it is important health workers can access and implement these predictive models in their clinical practice This means deploying a prospective tool across all mobile platforms with seamless offline functionality as health facilities may have compromised internet connectivity. The primary aim of this paper is to describe the deployment of a neural network model; a complete description of the machine learning algorithm is given in another publication [6]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
