Reduce patient hospitalization – through careful triage at their home or on arrival at the emergency room – is a particularly important goal during pandemics, when the continuous influx of ambulances can overload hospitals. A solution that, to tackle this problem, leverages artificial intelligence is now in phase 2 of experimentation at the emergency room of the IRCCS San Matteo Foundation in Pavia. It is a software platform, “Alphabet”, created by Laife Reply, which integrates the patient’s health information with the results of radiographs and other clinical tests to assist the doctor in predicting the evolution of the infection, so as to support in real-time triage decisions. The project follows Lorenzo Preda, director of UOC Radiology of the Policlinico San Matteo of Pavia and full professor of the University of Pavia.
Covid vaccine. Valle d’Aosta primacy: an algorithm decides who gets to first
by Francesca Soro
Professor Preda, what is “Alphabet”?
“It is an artificial intelligence system that we developed, with the support of Reply, during the first pandemic phase. It aims to integrate information derived from the radiological examination – also potentially carried out at the patient’s home or in the emergency room – with a series of clinical and anamnestic data (such as age, sex, comorbidities) easily available also from the electronic health file, with the involvement of the general practitioner. Integrating this information through this artificial intelligence tool allows an evaluation as objective as possible of the pathological condition of the patient. This has the dual purpose of providing objective, timely, quantifiable data on the clinical status at the time of the patient’s presentation, and also has the ambition to provide predictive information on the possible short-medium term trend of the pathology, infection, thus allowing to rationalize the hospitalization of patients: a hospitalizing, in particular, in hub centers, only patients who really need it. And instead treating patients in less severe stages of the disease at home or in peripheral centers (the so-called “Spoke”). And perhaps by identifying those cases that can be managed immediately with home assistance but which are at potential risk of developing a possibly more unfavorable evolution in the short to medium term. This system allows the territorial hubs to be pre-alerted on the number of patients who really need hospitalization “.
Coronavirus, biomarker discovered that ‘reveals’ the severity of the disease
by Daniele Banfi
However, it is not a system that replaces the doctor …
“Absolutely not. For example for the radiological part, it is clear that the evaluation of the examination needs an evaluation by the radiologist. The instrument can create a sort of priority degree in the evaluation by the radiologist in which normal radiograms, free of alterations, they are not attentive, while the cases in which the instrument highlights the most important radiographic alterations are. And the role of the radiologist is to confirm the severity and confirm that these alterations are compatible with interstitial pneumonia and therefore with the possible Covid infection. Because there are also other pathologies where the picture is related to heart failure or other possible causes “.
How is this second phase of the experiment going?
“In the first phase of development – in collaboration between the University of Pavia, Maugeri Clinical Institutes and San Matteo Polyclinic – all the cases were recovered anonymously. Now the second phase that we are launching in these days, having passed the phases of approval of the ethics committee, aims to try to apply this tool prospectively in the population of the new wave by verifying and validating the validity of the tool itself.In this second phase the idea is to enroll patients without using the tool for clinical use. The data of incoming patients are anonymized and sent to “Alfabeto” which produces its evaluations, which are not used for clinical use “.
Never confuse Covid and flu
by Fabio Di Todaro
So even in this second phase the doctor’s decisions are not made on the basis of the information given to Alfabeto?
“No, the information is simply collected. And then, having collected a series of at least 200 patients, we will check if the evaluations made by the instrument correspond to what will be the real evolution of the pictures. This will make us understand if the use of Alphabet would have improved the management of Covid patients. An interesting thing is that being Alphabet based on machine learning, it is also able to continue learning from new data. So this tool can also, as the pandemic evolves and the virus also mutates, continue to self-learning and therefore, potentially, improving performance by adapting to the evolution of the characteristics of the epidemic “.
What kind of data is the artificial intelligence system trained on?
“He was trained on anamnestic data, that is, sex, age and comorbidities. Because it was seen that the presence of pre-existing pathological conditions, such as chronic lung diseases or heart disease, represent a negative prognostic factor with respect to the evolution of the disease. Then the system incames. data relating to the patient’s medical history also deducible from the health file. Then the data derived from the chest X-ray taken at the patient’s bed, or in the emergency room. And some clinical parameters such as the degree of oxygen saturation, the feverish state or less than the patient, and some laboratory parameters readily available with blood samples “.
Longform
Covid: when to go to the hospital
by Tiziana Moriconi
How is this data processed?
“All this data is sent via the cloud to the virtual platform of Alfabeto, which brings together this information and produces a sort of” score “of the severity of the disease. And of the possible evolution of the picture. From a graphic point of view, in the validation phase and when the instrument is refined, a “traffic light” – a gradation of severity of the disease – will be seen with three degrees: red, yellow or green. To predict the possible evolution of the picture as well as to give precise information on the pathological situation “.
Is this system generalizable to other types of disease?
“Surely this platform and other similar tools that are set on a specific pathological condition, then can be applicable to other contexts. For example, the part of the evaluation of the radiological image was initially developed by Reply for a study conducted by the Maugeri clinical institutes on evaluation of mammograms “.
Covid: let’s buy an oximeter
by Valentina Arcovio
.