Artificial intelligence helps in the analysis of health exams in up to 15 minutes

Artificial intelligence is already in people’s daily lives, in the algorithm for selecting movies and series on digital platforms, in the selection of “personalized” advertisements and content on smartphones, in accessing security places such as airports and others.

Now, the analysis of image exams by artificial intelligence (AI) is also much closer to our daily lives — and that of Brazilians.

Officially launched across the entire network of diagnostic medicine laboratories in March this year, but already used experimentally in some centers since April 2019, Fleury has just incorporated the use of an AI tool called Aidoc into the network’s imaging exams.

Developed in Israel, the program can detect in up to 15 minutes diagnoses considered critical in chest tomography exams, such as the formation of clots that can cause a pulmonary embolism. The procedure does not replace the medical evaluation, but it helps in organizing the flow of exams performed to help the technician in charge to quickly identify a change and communicate the medical team.

With the application of artificial intelligence, the laboratory says it has already managed to save 656 lives. “We had already been testing the program for two years and the decision to expand its use to routine use in the laboratory was because we really saw a very big benefit, patients who had an early diagnosis and treatment because of AI”, explains Bruno Aragão Rocha , medical innovation coordinator at Grupo Fleury.

The expansion of the use of AI in all laboratories in the network is also accompanied by an expansion of the exams in which the tool is offered. In addition to chest tomography, the program was also implemented for skull tomography exams, to help in the identification of intracranial hemorrhage (as, for example, in the diagnosis of stroke), of the cervical spine for fractures in the neck region, and of abdomen, for intestinal perforations.

For Edgar Rizatti, Executive Medical Director of Grupo Fleury, the time saved with the use of AI in imaging exams allows for a much faster start of preventive treatment. “For example, if a thrombus is identified [coágulo]which is something that AI can identify well, the fact that it saves this time the doctor can start the anticoagulant, preventing this thrombus from increasing in size and causing a more serious accident”, he explains.

It is not the first time that AI has been used for health diagnoses. In addition to the detection of cases of stroke, it was already being applied in the analysis of magnetic resonance imaging, in radiology exams for the detection of lung cancer, breast or bone fractures and even in the assessment of fetal health in pregnant women.

In the case of the latter, a group of Brazilian scientists has used AI to assess the occurrence of problems during pregnancy that can lead to some type of medical condition in the baby, such as heart problems.

Last Thursday (7), a study published in the journal Nature Cardiovascular Research showed that the use of AI could predict the risk of arrhythmia infarction. The research, led by scientists from John Hopkins University (JHU), in the United States, developed a tool capable of finding small scars in the heart muscle that may indicate the occurrence of minor cardiac arrests. These scars are virtually invisible to the naked eye, even in high-resolution imaging tests such as MRI.

Natalia Trayanova, a researcher at JHU and one of the authors of the study, told Sheet that the tool used was able to accurately predict the risk of a potential infarction in 156 patients analyzed, obtaining greater efficacy when compared to the control group, which consisted of the analysis of patients by human physicians in 60 hospital centers.

“We hope that our program, called SSCAR [acrônimo para estudo de sobrevivência de risco para parada cardíaca arrítmica] can be adopted in other hospital centers to bring benefits to many patients”, he said.

The accuracy, however, varies enormously from database to database, explains Rocha, from Fleury. “It’s never going to be 100%. And it’s interesting that, in cases where the AI ​​fails, they are different from the cases where the human eye fails. Also, the tool is influenced by the population it was based on, in the tests that were shot.”

In the last March 29 issue of the scientific journal Radiology, two studies were published comparing the use of AI with the medical report. The first, a systematic review of the use of AI for diagnosing bone fractures, found an accuracy (ability of the method to get the diagnosis right) of 92% for AI against 91% of physicians, and a specificity (probability of the test being negative, since the individual does not have the disease) of 91% for AI and 92% for humans.

In the other study, which analyzed about 122,000 mammography exams from October 2009 to December 2018, the use of AI was able to identify a high risk for breast cancer in 77.9% of exams with a confirmed diagnosis of cancer, of of which 86.8% were detected in clinical screening and 44.9% who initially received a negative diagnosis but were positive for the presence of the tumor.

Less than 1% of the cancers that were diagnosed by doctors were given a risk of 1 by the AI, that is, it was not able to identify these cases as a potential tumor.

Because of this, experts say that AI should not replace humans, but rather be an auxiliary tool in diagnosis. “Analyzing an exam or interpreting a change is more complex than just identifying a little ball, a specific stain. Therefore, AI is complementary in some tasks, but the time will come when it will replace the human”, says Rocha.

“Today, the volume of data we work with, whether from laboratory tests or imaging, is more than astronomical. We had already been using bioinformatics tools to treat these images, and in some areas the practical application of AI can bring a clear benefit, in addition to the time, faster results for patients”, completes Rizatti.