Computers can learn to interpret even the smallest variations in myocardial status
Text: Jaakko Kinnunen
Photo: Jenni Toivonen
A recent study conducted at the University of Tampere in Finland showed that machine learning can be used to diagnose genetic heart diseases. The study used artificial intelligence to compare cardiomyocytes from CPVT, long QT syndrome (LQTS) and hypertrophic cardiomyopathy (HCM) patients. The results were published in the prestigious Scientific Reports of Nature Springer.
The researchers used algorithms to compare how cardiomyocytes store and release intracellular calcium in different diseases. Calcium metabolism is important for the healthy functioning of the heart. Previous studies had already shown that the calcium metabolism of healthy cardiovascular cells differs from that of diseased cells. Cellular arrhythmias or small variants in calcium transients may be used to diagnose heart conditions.
Professor Katriina Aalto-Setälä says that the results surprised the researchers, too.
“The results were unbelievably good. It is really astonishing that artificial intelligence was able to detect heart diseases so accurately,” Aalto-Setälä says.
Professor of computer science Martti Juhola represents expertise in machine learning in the project. Researchers Henry Joutsijoki and Kirsi Penttinen also participated in the study.
“Our study consisted of two parts. We first identified the peaks in the patient’s calcium transient signals. This was followed by classifying the diseases, in which artificial intelligence played a significant role. Henry Joutsijoki did a great job there,” Juhola says.
The study was conducted by using calcium transient signals derived from video images of pulsating heart cells; they indicated calcium storage and release in the body. Artificial intelligence, i.e. machine learning methods, learned to detect minuscule variations in the pulsating cells, which showed the presence of different diseases.
“People would not be able to do what the computer did because the variations were so tiny. People are good at identifying other things, but in this kind of work people cannot beat computers,” Juhola points out.
The accuracy of the predictions made by artificial intelligence depends on the size of the data the system is able to use.
“In this study, we were very lucky because the diseases we compared were clearly different from one another. That is by no means self-evident,” Juhola says.
The results are very promising for the researchers because the computing power of machines is steadily growing and increasing amounts of high-quality patient data are available.
According to Juhola, a new study is already under way to investigate new heart diseases together with the three previously analysed ones.
“The goal is ultimately to create a clinical application that will significantly benefit doctors and patients,” Juhola says.
Human induced pluripotent stem cells (iPS cells), which are able to differentiate into the cells of any tissue, were used in the study. Kirsi Penttinen did the demanding cell differentiations and functional analyses of cardiac cells. The cardiac cells used in the study were generated from the patients’ skin cells.
The researchers hope that induced pluripotent stem cells and artificial intelligence will make diagnosing more efficient in the future.
“If a patient presents unspecific symptoms, we may use iPS cells and artificial intelligence to detect the disease the patient is suffering from. In other words, we have the possibility to target these analyses to specific cases,” Aalto-Setälä says.
The long QT syndrome (LQTS) is a disease that exposes patients to severe arrhythmias. Most of the patients who carry the gene causing the disease are asymptomatic, but the disease may also cause a sudden loss of consciousness or death.
The symptoms of CPVT – catecholaminergic polymorphic ventricular tachycardia – include stress-induced arrhythmias that may cause unconsciousness and sudden deaths.
Hypertrophic cardiomyopathy (HCM) is a disease where the heart muscle has thickened. In HCM, the heart pumps blood normally, but blood flow to the myocardium has deteriorated. The disease may result in sudden death.
Please see the article in full:
Juhola Martti, Joutsijoki Henry, Penttinen Kirsi & Aalto-Setälä Katriina: Detection of genetic cardiac diseases by Ca2+ transient profiles using machine learning methods. Scientific Reports, volume 8, Article number: 9355 (2018)