Artificial Intelligence – Sars Covid Risk Evaluation
AI to fight Covid-19 emergency
7 December, 2020
Reading time: 4 minuti
The project started during the first phase of the pandemic, with the aim of using artificial intelligence to solve one of the main problems encountered during the first months of the health emergency: to predict and calculate the risk of mortality for Covid-19 starting from the clinical data of positive and hospitalized patients in order to correctly and quickly address their therapeutic path.
At this time we are in the final stages of release and integration with the information systems of the San Raffaele Hospital in Milan, in order to make it easily accessible and usable by medical staff. In this way the AI system becomes a tool daily using during this second phase of the pandemic and at the same time we have the possibility to validate the model in production, with the data of patients currently hospitalized and positive to Covid-19.
“Research is a fundamental key to fight Covid-19, to increase the knowledge of the virus, its effects and the ability to respond to future emergencies” the words of the Vice-President of the Regione Lombardia open the press release on the allocation of funds for the winning projects of the call for 7.5 million euro “Covid-19: insieme per la ricerca di tutti” that has united the Regione Lombardia, Fondazione Cariplo and Fondazione Umberto Veronesi.
The founded projects are 27 and they involve Universities, Local Hospitals and Companies. One of these is AI-SCoRE, an acronym for Artificial Intelligence – Sars Covid Risk Evaluation, which has seen us working closely with the University Vita-Salute San Raffaele, ASST Bergamo EST, Centro Cardiologico Monzino Spa and Porini Srl, with the contribution of Microsoft and NVIDIA.
The first phase of the project has seen the commitment of 15 Hospitals in Northern Italy severely affected during the first months of the pandemic that have made available clinical, demographic and radiological data about 1800 patients. The data were analyzed to assess quality and consistency and to identify the parameters that most influence the patient’s hospital progression. They also were used to train different models of machine learning and artificial intelligence, in order to identify the most suitable one for estimating the risk of mortality starting from the parameters detected when the patient was admitted to the hospital. The logistic regression model obtained the best results, with a performance comparable to similar studies recently published. Finally, we developed a Bayesian version of the model to allow a better interpretability of the prediction. In this way it will be possible to associate a degree of uncertainty to the calculated risk value, an important element for the doctor to keep under control the judgment of the artificial intelligence system that supports him in his work practice and to make timely decisions based on verifiable information.
The innovative aspect of the project is related to the practical use of the models in hospitals, made possible by the technological platform developed ad hoc, in compliance with the regulations of software as a medical device and according to the privacy law. Through the infrastructure deployed, we are able to offer Hospitals the execution of the models in the required time, complete traceability of data and models (through a versioning system) and real-time monitoring of performance. Moreover, through active learning logics the models will be automatically improved day by day, based on the new data available so that the predictions become more and more accurate and useful to the doctors who will use them to define the best therapeutic path for the specific patient.
In March, the MIT Technology Review made this title: “AI could help with the next pandemic—but not with this one”.
The author, Will Douglas Heaven, cited famous systems that use artificial intelligence to monitor and predict complex phenomena such as pandemics, wondering why they did not work in the case of Covid-19. He came to the conclusion that
“AI will not save us from the coronavirus—certainly not this time. But there’s every chance it will play a bigger role in future epidemics—if we make some big changes. Making the most of AI will take a lot of data, time, and smart coordination between many different people.”
We believe that the AI-SCoRE project can be a practical example of how to approach a complex artificial intelligence project, which aims to far beyond the replication of the diagnostic activity of a single specialist, but which aims to use these techniques to obtain the overview that only a team of doctors could have in times and ways not compatible with the health emergency we are experiencing. The models and the platform implemented are intended to be a tool in the hands of doctors to manage the current situation and to use the large amount of information and experience gathered in recent months to create new knowledge useful for future emergencies management.
Want to learn more about other projects?
Visit our case histories section.
Metallurgy: aluminium die-casting from data to action (and savings). An interview with Alberto Albertini - Innovation Center Director, Technology Scouting at Antares Vision Group