St Petersburg doctors teach neural network to detect the risk of complications for patients with heart disease
Cardiovascular disease remains the most frequent cause of death in Europe, despite significant advances in diagnosis and treatment. Early and accurate diagnosis and prognostic assessment can help to reverse this situation. Andrei Obrezan, Professor, Head of the Department of Hospital Therapy at St Petersburg University, Chief Physician of the SOGAZ MEDICINE group of clinics, and Timur Abdualimov, Cardiologist at the SOGAZ International Medical Centre, suggested using neural network, a machine learning method modelled on processing large volumes of information.
The article ‘Prediction of the fact and degree of coronary artery disease using the processing of clinical and instrumental data by artificial intelligence’ is published in the journal ‘Vestnik of St Petersburg University. Medicine’.
Doctors used a comparison of results of coronary angiography, a procedure now considered the gold standard in diagnosing the condition of heart vessels, and electrocardiogram data to train the network and prepare it to analyse real patient cases. Artificial intelligence was tasked with studying the diagnostic-relevant parameters of more than 100 patients between the ages of 31 and 89. The goal was to learn how to classify coronary arteries, detect the presence of vascular lesions and predict the occurrence of coronary disease, i.e. blood circulation disorders in the heart muscle.
The physicians used neural network to analyse information on 130 patients from the test group who had undergone elective or emergency coronary catheterisation. Their medical records, including age, gender, diagnosis, pathology, presence or absence of related diseases, heredity, bad habits, and electrocardiogram results, were uploaded into the machine learning database. Artificial intelligence studied the data and determined which patients would face main coronary artery lesions and coronary heart disease.
The doctors also examined all the patients tested by conventional means. To predict a complicated course of coronary artery disease, they were examined by computerised coronary angiography (a study of the heart vessels by injecting an X-ray-contrast agent into the arteries) as well as daily electrocardiogram monitoring and a treadmill stress test to assess heart performance during physical activity.
The study showed that neural network performed better than traditional diagnostic methods. For example, artificial intelligence achieved 93% accuracy in detecting myocardial ischaemia, while daily ECG monitoring was only 87% accurate.
‘The process of medical decision-making is a complex activity. It is based on the availability of objective and reliable evidence, access to knowledge, and correct interpretation of the available data, taking into account the risk-benefit ratio for the patient. Long-term prognosis plays an important role in treatment of cardiovascular disease. The results of our tests have proven the high potential for practical application of machine learning methods in clinical practice,’ said Andrei Obrezan.
Work on introducing neural network into medical practice will be continued.