Coronavirus Disease 2019 (COVID-19) can lead to acute respiratory distress, multiple organ failure and ultimately death. Optimising patient care in the hospital is therefore of utmost importance. Artificial Intelligence (AI) in the form of machine learning paired with rigorous mathematical and statistical techniques can help in this. By processing and analysing the complex, multi-stream patient data that is collected routinely in the hospital, these methods can predict clinical phenotypes of the disease. Front-line clinicians can use this information to assist their clinical decision making and also for resource allocation, with great potential to save lives in the process.

AI offers clear potential for the analysis and interpretation of large multi-stream datasets curated on COVID-19 in hospitals. In particular, it has increased sensitivity and specificity of diagnosis and prognosis by drawing connections from an individual patient to a large cohort and associated disease phenotypes and disease progression, the ability of integrating a large number of different and potentially complex types of data, e.g. imaging, clinical data, laboratory data, to aid healthcare decisions at the bedside.

In this project we will address the key bottlenecks for unleashing the potential for COVID-19 AI-healthcare support, namely:

  • the bias in small data and the variability in large internationally-sourced data sets,

  • the integration of multi-stream data, in particular chest-imaging and clinical data,

  • the difficulty and complexity of severity prediction and prognosis,

  • the urgent need for clinicians and data analysts to work side-by-side to ensure the developed AI tools are rapidly implementable in the clinic.

On 23 March 2020 a team of mathematicians, engineers, radiologists and clinicians gathered to develop an AI assisted diagnosis and prognosis tool for COVID-19 that could support patient management in the hospital. We aim to develop an open-source AI tool that integrates multi-stream data to support the diagnosis and triaging for COVID-19 in the UK. Our tool will be accompanied by a well-documented algorithmic strategy that will allow implementation and tuning to other countries.

We have curated a large dataset of imaging and clinical datasets from Austria, Brazil, China, Italy, Russia, Spain, the USA and the UK for developing, testing and validating our algorithms. Our developed AI tool is based on state-of-the-art deep learning approaches combined with rigorous mathematical and statistical techniques that are customised to the needs of COVID-19.

It is crucial to recognise that this is just the first pandemic of the machine learning age and we must be better prepared in future. We are preparing a blueprint for how we should respond next time to allow rapid data collection, fast algorithm development and robust assessment of algorithm generalisability.