In recent years, we have heard quite a lot the term Artificial Intelligence (AI). Not only in the field of technology but also in the area of business and everyday life. Cars that are parked by itself, phones that answer questions kindly, personalized attention to customers through chatbots, are some of the examples in which we see that Artificial Intelligence.
Interestingly, you can use this type of technology to identify heart failure potential.
Artificial Intelligence Predicts Death from Heart Failure With 73% Accuracy
A scientific team has created a system based on Artificial Intelligence (AI) that has learned to predict with high precision the risk of death in patients with cardiovascular disease.
The platform was able to check for lethal heart attacks in patients with pulmonary hypertension with an accuracy of up to 73%. The researchers’ work was published in the journal Radiology.
The predictions of the developed system were made possible by the very nature of the disorder itself. Pulmonary hypertension is characterized by an increase in blood pressure in the pulmonary artery and a greater blood load in the right ventricle of the heart. This leads to the development of heart failure and, thus, to premature death.
To determine to what level of risk of heart failure patients are exposed, doctors make complex diagnoses: magnetic resonances of the heart, electrocardiograms, ultrasound, X-rays of the lung, among others.
However, it is not always possible to accurately predict the probability of deterioration of patients.
How does it work?
Through machine learning – a technique developed to allow AI systems to learn on their own from massive amounts of information, in a series of tests and errors – the platform designed to analyze the cardiac examinations of 224 patients, to predict the probability of death of the patient.
On this basis, the program constructed a three-dimensional model of the body and measured the movement of 30 thousand individual points in it during the span of each heartbeat.
This data analysis was then contrasted with the medical records of the patients’ health over eight years.
The group of scientists tested the program on 32 patients. The computer was able to identify among these those who were going to live at least one more year, with up to 73% accuracy.
The doctors also had images of the heart, hemodynamic data and clinical markers could only predict the same with an accuracy of 60%.
The authors of the paper assure that their system can build a forecast for the next five years.
To address this challenge, Scientists and doctors used an automated learning classifier integrated into a tablet or smartphone to record 17 objective dichotomous variables, physical examination data and subjective variables of patients with precordial pain of unknown origin during the triage: age, weight, height, cardiovascular risk factors, systolic blood pressure, heart rate and characteristic, intensity, and location of pain, among others.
Future Application
At the moment, scientists – whose work took a total of four years – plan to test the system in patients from different hospitals.
In the future, the program could be used to diagnose other types of cardiovascular diseases, which lead to heart failure, such as cardiomyopathy.
Last December, a group of American scientists used magnetic resonance imaging and machine learning with AI to issue early diagnoses of melanoma.
According to the authors, the study is based on the fact that chest pain represents one of the most frequent causes of medical emergency services, and that interpreting the information obtained from the history.
And objective data depends greatly on the experience of the doctor in charge: the review of the evidence shows that the “atypical” symptoms do not allow to rule out acute coronary syndrome, while the “typical” ones do not enable it to be confirmed.
The artificial intelligence tool used consisted of a type of supervised machine learning algorithm called random forest or random forests of 150 estimators, which combines multiple algorithms of the same kind to generate a more powerful predictive model.
After a first phase of learning by entering known data, the classifier can create modeled knowledge that provides predictive output data.
