Notícia

Deans' Stroke Musings

From diagnosis to rehabilitation, artificial intelligence helps stroke patients

Publicado em 27 setembro 2021

Artificial intelligence is increasingly being used in diagnosis, treatment and rehabilitation of patients after a stroke (stroke), the world’s leading cause of disability, removing one in six people from their activities.

THE Federal University of São Paulo (Unifesp), for example, studies applications to track and predict complications. One of them is the transformation of ischemic stroke (blood vessel obstruction) into hemorrhagic (blood leakage). From the collection of data from more than 3,000 patients from several Brazilian centers and one American (Columbia University), the research became a reference in several countries and was awarded by the Brazilian Academy of Neurology.

Another study addresses the complication of late cerebral ischemia in patients with subarachnoid hemorrhage (a condition usually associated with ruptured cerebral aneurysms). “The idea is that these data (pressure, temperature, heart rate, laboratory changes and transcranial Doppler data) generate a model that alerts us days in advance about the risk of complications”, says researcher João Brainer, co-advisor of the Postgraduate program -Graduation in Neurology and Neurosciences at Unifesp. The university is also researching, in an unprecedented way in the world, the identification of dysphagia (difficulty in swallowing food) based on voice recognition.

These researches try to minimize dramas such as that of the billmaker Sandra Schulze, who had an ischemic stroke in September 2013. After a heavy night’s sleep, she, 52, fell right out of bed. He could no longer move, his left arm cold and paralyzed. There were three days in the ICU, 15 in a hospital in Joinville. She is still in physiotherapy, but she is grateful to the professionals who helped her from her first days in the hospital.

The rehabilitation of patients, in fact, is the focus of a pioneering study by the Oswaldo Cruz do Ceará Foundation with virtual reality. Wearing glasses that create a parallel reality and guided by a healthcare professional, the patient simulates daily activities. Each movement is captured by sensors, which form a database to assess and monitor the progress of the treatment. It will also be possible to accurately quantify information from space, movement speed, limb angulation and action precision. This new technique will begin to be clinically validated in October at the Hospital Geral de Fortaleza, which treats more than 1,900 stroke cases per year.

How it works

Applications, robots, algorithms and software that configure artificial intelligence perform analyzes that human beings cannot. Doctors typically assess two to three dimensions, such as the outcome of a patient’s examination over time. When the possibilities increase, the human brain has difficulty processing. This occurs in the case of genetic variants and drug interactions, for example. There, the machine helps to reinforce the diagnosis.

Systems are fed a “fundamental truth” – or a basic pattern from which other decisions are generated. If machines are taught that a particular pattern displayed in the image is a brain tumor, every time the same abnormality is seen, the system will label it the same way.

But how do they do it? This is one of the questions proposed by the Artificial Intelligence Center (C4AI), a partnership between FAPESP, IBM and USP. One of the lines of research is related to “learning” algorithms. “Algorithms can ‘see’ things that escape us. We want to know exactly how this happens”, explains José Krieger, physician, researcher at Incor and one of the project leaders.

Another research by C4AI studies the modeling of strokes. From electroencephalograms (EEGs) from the Neuromodulation Laboratory of the Physical Medicine and Rehabilitation Institute of the Hospital das Clínicas of the USP Medical School, the researchers developed a stroke classification system with machine learning techniques, i, set of techniques that can be turned into algorithms. Marco Antonio Gutierrez, IT director at Incor, reveals that a database with 200,000 ECG treatments from patients was used. “After the specialists made the reports, we developed an algorithm to learn how to look for the same information in the treaties that the specialist looks for.”

success stories

Artificial intelligence is not yet part of the Brazilian hospital reality. The Institute for Research and Study for Diagnostic Imaging Foundation (Fidi), which manages radiology and diagnostic imaging in the public sector, is present in 76 hospitals. Of this total, only eight have artificial intelligence technology, seven in São Paulo and another in Goiás. “Usage is not yet the rule. There are few ‘commercial’ algorithms, approved by Anvisa. In addition, there is the challenge of implementation, with integration between different systems”, says radiologist Igor Santos, Superintendent of Innovation at Fidi.

“We’ve come a long way in the last five years, but not all hospitals use the treatment (with the help of artificial intelligence)”, says vascular neurologist Sheila Martins, founder, president of the Brazilian Network of Stroke and professor of Neurology at the Federal University of Rio Grande do Sul (UFRGS).

Neurologist Octávio Pontes, a professor at the Faculty of Medicine at USP-Ribeirão Preto and coordinator of the National Network for Research on Stroke, says that costs are a limitation. “It is important to partner with companies with products in development to incorporate these advances in diagnosis, prevention, treatment and rehabilitation of patients.”

Those who already use algorithms count advances. The Hospital Israelita Albert Einstein, which has been using the technology since 2019, points out that some benefits are the diagnostic aid to the neuroradiologist, the automated quantification of areas of ischemia and potentially salvageable with the treatment, as well as the greater resolution of the tomography in cases of patients who wake up with a stroke or who are outside the 6-hour therapeutic window.

At the Hospital de Clinicas de Porto Alegre, as Sheila Martins reports, one of the achievements is the use of perfusion software, capable of showing areas of the brain that have already died (pink) and those that are in pain, but which could be saved (green). “This is essential in cases where the patient arrives late at the hospital or it is not possible to specify the time when symptoms start. This allows treating more patients”, he says.

In the opinion of Igor Santos, one of the main advances is the streamlining of service. “The normal period for an emergency exam, whether in public or private, is two hours. Time between taking the exam and receiving the report. With artificial intelligence it is possible to reduce this to 30 minutes. This is fundamental”, calculates Igor Santos.

This time was crucial for the patient Antonio Valentim da Silva. The 71-year-old retiree complained of dizziness, headache and neck pain in early September. It was a stroke. He was admitted to Hospital Mandaqui, one of the hospitals with Fidi’s presence, located in the north of São Paulo, at 10:31 am on the 7th. At 10:59 am, he underwent a CT scan using artificial intelligence and the report was issued at 11:44 am. He says he is fine, but takes medication and should return to the doctor within three months. It has no sequelae. His wife, Arcendina Cândida, says that the speed of service was essential. This Friday, she and her husband completed 44 years of marriage.

Nelson Fortes, coordinator of diagnostic neuroradiology at Hcor, one of the pioneer hospitals in the use of technology, emphasizes the importance of combining the importance of human intelligence with artificial intelligence. “The machine does not replace the human being. She guides the clinician,” he explains. “In cases of stroke, in which the patient is restless, the movement can generate a false positive.”

Day to day

At Albert Einstein Hospital, the protocol for using artificial intelligence is as follows:(This seems way too slow and complicated. After doing all this, what are the chances of 100% recovery? That is the only endpoint in stroke.)

1. Arrival of the patient suspected of having a stroke at the hospital;

2. Evaluation with a neurologist and performance of cranial tomography and arterial tomography angiography;

3. Connected to the tomograph, Artificial Intelligence (AI) software assesses the presence of a stroke and, if so, its extension;

4. In the CT angiography performed together with the tomography, the AI​​analyzes the presence of vessel obstruction and identifies the percentage in which the arterial flow is reduced;

5. In selected cases, a cerebral perfusion study is performed, another imaging exam, where the AI ​​quantifies the areas of injury and those at risk of ischemia;

6. The neuroradiologist analyzes the images taken and the post-processing results and communicates the final result to the neurologist;

7. In selected cases, magnetic resonance imaging can be performed; analyzes of infarct size and areas at risk for ischemia are also evaluated by the AI;

Health