ITMO: Researchers Create an Algorithm To Help Reduce the Risk of Complications in Cardiovascular Diseases

The machine-learning-based algorithm is developed by ITMO’s . By analyzing data on patients’ well-being and lifestyle, the program will help physicians make timely decisions on prescribing treatment, drugs, or surgery.

Cardiovascular diseases remain a major threat to humanity – they are the cause of death in the world. Moreover, the death rate of heart disease keeps growing, while strokes and heart attacks are getting “younger.” For instance, working-age Russians die from cardiovascular diseases eight times more often than Norwegians in the same age group.

The development of cardiovascular diseases is hard to predict. Furthermore, they are linked to a great variety of factors, such as one’s diet, environment, bad habits, emotional state, sleep routine, or working conditions. All of these factors combine to make the work of a cardiologist especially challenging.

Working with many unknown variables
For decades, cardiologists have had to base their decisions only on blood tests, which rarely reflect the whole picture. Naturally, experience and intuition were also of help. These days, there is a rising demand for computer algorithms that would help physicians make decisions for patients whose lives literally depend on them.

However, such programs have their drawbacks – their answers are hard to interpret, while some of them cannot be trained very well on a limited amount of data. Moreover, it’s hard for algorithms to make predictions for nonlinear interactions, or, in other words, those factors that indirectly affect the development of cardiovascular diseases. Finally, algorithms and the manuals on how to create them are not as accessible for cardiologists as desired.

Ilya Derevitskii, an engineer at ITMO’s National Center for Cognitive Research, has created an algorithm devoid of several drawbacks of similar medical programs. The algorithm was developed in collaboration with the Almazov National Medical Research Center.

Thyroid atrial fibrillation
The algorithm helps predict development of thyroid atrial fibrillation, which is cardiac arrhythmia caused by the toxicity of . As a result, because the spontaneous electrical impulse that makes the heart beat is distrurbed, the patient’s auricles don’t contract in sync with each other.

Graph of thyroid atrial fibrillation development. Illustration by Ilya Derevitskii
Graph of thyroid atrial fibrillation development. Illustration by Ilya Derevitskii

“It means that the heart cannot fulfil its function and enrich the blood vessels with the amount of blood required at a given moment,” explains Daria Ponomartseva, an endocrinologist at the Endocrinology Department 1 of the Almazov National Medical Research Center. “This leads to cardiac insufficiency. Moreover, the blood flow inside the auricles slows down, which can lead to blood clots inside the auricles’ cavities. Thyroid atrial fibrillation is demonstrated to be linked to higher mortality and lower quality of life. However, if it hasn’t been developing for a long time, it’s a reversible disturbance of the rhythm. In some patients, atrial fibrillation stops when hyperthyroidism is cured.”

The algorithm created by Ilya Derevitskii, combines the results of classic machine learning algorithms with dynamic analysis methods for better modelling and prediction of the disease’s development. The algorithm allows viewing the illness as a succession of changes in the patient’s well-being and thus predicts its course.

The system is created to take into account both test results and indirect factors that can affect the course of the illness. For instance, the algorithm analyzes the patient’s gender, age, height, weight, body mass index, causes of hyperthyroidism, smoking status, and cholesterol level. This data is juxtaposed to the information about the patient’s potential hypertension, coronary artery disease, chronic cardiac insufficiency, or other illnesses. The algorithm also takes into account characteristics of hyperthyroidism and any concomitant cardiovascular diseases.

“Practical results include a survey scale that evaluates the risk of the illness. We have also developed a method to predict the type of the developing thyroid atrial fibrillation,” says Ilya Derevitskii.

Part of the survey that evaluates the risk of thyroid atrial fibrillation. Illustration by Ilya Derevitskii
Part of the survey that evaluates the risk of thyroid atrial fibrillation. Illustration by Ilya Derevitskii

“This information will help identify patients with a high risk of thyroid atrial fibrillation or those with an unfavourable type of this disease. These patients will be treated more attentively, they would have to pay more visits to physicians, make more electrocardiograms, and take more hormone tests in the course of their treatment. Earlier radical treatment is meant to prevent the development of fibrillation. It may be that for patients with an extremely high risk of fibrillation, we would have to proceed to radical treatment at once, without the longer conservative treatment,” adds Daria Ponomartseva.

Open code
The algorithm is now in development and is undergoing various tests. It gets better by using more complex and precise algorithms, as well as new datasets from other clinics and hospitals. When this process is complete, the method is planned to be implemented in medical practice.

“The hybrid approach to predictive modelling of treatment is crucial for improving the quality of life and longevity of patients. In the near future, I am planning to use imitation modelling, such as discrete-event simulation, system dynamics modelling, and neural networks, to upgrade the algorithm. This will help predict the whole “health trajectory” of a patient for decades to come, including the sequence of important milestones in terms of the course of the illness. This solution will be available for healthcare specialists in the form of special software,” shares Ilya Derevitskii.

Comments are closed.