Artificial intelligence is support in the ICU and points out a serious case – 7/1/2023 – Tech

Algorithms used to prioritize content from friends on social networks, indicate series and movies or the best transit route can also, in healthcare, help to monitor data from patients hospitalized in ICUs, predict the availability of beds and medicines and give alerts in health exams. image about possible serious diagnoses.

In recent years, projects involving artificial intelligence —technology that seeks to imitate or reproduce human intelligence and our learning process— began to appear in some hospitals and clinics in the country, most of them connected to the private network and in large centers.

Use, however, is still limited compared to the potential of the area, according to experts, who predict a transformation in the coming years.

“We are still in the prehistory of artificial intelligence in health”, says Alexandre Chiavegatto Filho, professor of artificial intelligence in health at the Faculty of Public Health at USP.

Among the impasses are costs, implementation difficulties, lack of integration of some health data systems and lack of qualified professionals to work with these models. Another point is the need for a greater volume of tests compared to other areas.

The advancement of these technologies, however, is already seen as a certainty. “Health will most likely be the area most transformed by artificial intelligence. There is potential for the use of algorithms at all times during patient contact with the health system.”

Two reasons reinforce this analysis: the fact that the area collects and stores a large volume of data and the need to obtain answers to complex questions. “No one dies just because they eat poorly and are sedentary, it is always a complex interaction of factors, which makes algorithms have the potential to transform health.”

Data from the ICT Health survey, carried out by Cetic.br, indicate that, in 2022, at least 3,567 health establishments used AI — of these, 3,231 were from the private network. The total is equivalent to about 3% of the 120,000 health establishments surveyed.

According to sector representatives, use is more frequent in management support tools, such as systems that seek to estimate the length of stay of patients and help predict the demand for beds, or that focus on medicines and organization of queues in health .

Diagnosis support initiatives, however, have also advanced in recent years —especially in radiology.

At the Hospital das Clínicas da Faculdade de Medicina da USP, for example, one of the first experiences in this regard took place during the Covid pandemic, when the artificial intelligence team created, with the support of companies, an algorithm to identify pulmonary impairment in CT scans of patients with the illness.

“This was made available to dozens of hospitals, which helped to treat and save many patients”, says doctor Giovani Cerri, chairman of the boards of the Institute of Radiology and Innovation at HC.

Currently, the hospital also has artificial intelligence tests to support the diagnosis of other conditions, such as breast cancer, lung nodule and fractures.

It is also developing a platform that uses algorithms to screen liver lesions and tumors in imaging tests. The objective is to make the tool available to reduce the waiting time for results in the public network.

“The system will work as a triage of positive cases so that the report comes out faster. Today, in the public service, this can take a few weeks. With the system, it would come out in a few days”, says radiologist Bruno Aragão, researcher responsible for the project.

Developed since 2018, the system is now in the data validation stage. “We hope that in a year or two it can be used more widely,” says Cerri.

The examples go further and reach other functions.

At Rede D’Or, since September, a system crosses variables such as scheduling time, displacement, timetables and other factors to predict a higher risk of missing appointments and optimize the agenda.

“Our absentee rates were around 25% to 30%. We used the history of absent patients and created a model that was able to say with an 80% chance of coming to the consultation. This reduced absenteeism by 33%, because it acts to confirm or give another date option”, says Daniel Ferraz, doctor in charge of the artificial intelligence area.

The network is now betting on other fronts. In one of them, a natural language processing algorithm was created (a branch of AI that focuses on understanding human language) that reads medical records and test reports to help identify diseases such as breast cancer.

In another, a system analyzes images of the back of the eye to help triage patients with eye problems and refer the case to specialists. The proposal is in the testing phase at the São Rafael hospital, in Salvador.

Clinics that have already implemented the use of AI also report impacts. This is the case of Grupo Fleury, which has been using since last year a system capable of detecting, in up to 15 minutes, suspected cases of pulmonary embolism and intracranial hemorrhages in the analysis of CT scans.

According to Edgar Rizatti, president of medical, technical and hospital business units, the tool does not replace medical evaluation, but it helps to identify serious and urgent cases.

“Many times it is an accidental finding and occurs during the examination. The analysis identifies a thrombus in the large vessels of the lung, which can lead to an unfavorable outcome and lead to death”, he reports. “With this, the radiologist prioritizes the case for analysis and contacts the patient’s physician so that measures can be taken quickly.”

In 2022, around 24,000 tests were analyzed – of this total, at least 800 patients were referred after the alerts.

WHO warns of risk of errors

But if, on the one hand, the sector sees advances in the use of artificial intelligence, specialists also recommend caution in the incorporation of this type of technology in health.

“We need to be careful not to blindly accept everything that is labeled as artificial intelligence”, says Chiavegatto, from USP, who defends that new AI technologies in health undergo randomized studies, just as it happens for medicines.

The concern is in line with the warning issued by the World Health Organization in the wake of the launch of ChatGPT, a robot that uses artificial intelligence to generate responses in a few seconds.

“The hasty adoption of untested systems can lead to errors by healthcare professionals, cause harm to patients, undermine confidence in AI and delay the potential benefits and long-term uses of such technologies,” the organization wrote.

The position is shared by part of the doctors who accompany the area. Cerri, from HC, sees the potential for using models similar to ChatGPT as an alternative for contacting patients and facilitating research. But she has reservations.

“ChatGPT can give correct or very wrong information. It is not committed to the truth, only to provide information, which has to be filtered.”

In 2021, other aspects had already been alerted by the WHO in relation to artificial intelligence, such as patient data privacy and the risk of errors and biases by algorithms, which can accentuate inequalities.

A similar problem was identified by researchers from Unicamp who have been developing, since 2014, a project using algorithms for diagnosing melanoma. After achieving high precision in the analysis of photographs in initial studies, the group saw that the result was not repeated when observing possible skin lesions in the black population. The reason was the low number of images of this group in the database, obtained mainly from other countries.

“AI models learn from data and based on patterns. If you don’t have [dados da população negra], technology will not even see this pattern. And this is not the representation we have in Brazil”, says Sandra Ávila, a professor at the Institute of Computing who is in charge of the project.

Now, the group is doing new research to correct the problem and, in the future, be able to make the tool available to the SUS.

For Félix Rigoli, senior consultant at the core of studies on bioethics at Fiocruz Brasília, the inclusion of artificial intelligence in the SUS can help organize queues and facilitate care where there is a shortage of professionals.

“But this has to be incorporated in accordance with the principles of the SUS, that is, it must be egalitarian and not create differences. Which is not necessarily in the minds of everyone who invents artificial intelligence.”

There are still other challenges. “The first is to computerize. Today we have public units without or with very little internet connectivity”, he says. “Another issue is to increase the security of the database, which has already been targeted by hackers.”

Recently, representatives of the digital health area of ​​the Ministry of Health declared that the folder intends to include artificial intelligence in the SUS, but without a deadline.

One of the impasses would be the absence of a specific law on the subject, which leaves room for doubts regarding issues such as accountability in case of errors.

Currently, software for medical diagnosis with artificial intelligence is subject to the rules for registering medical devices at Anvisa. There is no specific balance on the subject.