Insights

Detecting heart failure with AI: HeartKinetics and Kantify

Collaboration between Kantify and HeartKinetics around real-time detection and classification of heart failure

HEART FAILURE

Heart failure is a chronic cardiovascular condition where the heart muscle is unable to pump enough blood to meet the body’s needs for blood and oxygen. Patients who experience an event of heart failure usually experience non-specific symptoms, such as tiredness, shortness of breath, and swollen legs. It is estimated that around 64.3 million people worldwide are living with heart failure. Treating an event of heart failure, especially in patients with a severe form of the disease, is critical to prevent both long-term organ damage and other cardiovascular risks.

As such, the timely diagnosis of an event of heart failure is crucial for patients suffering from the condition. However, diagnosing heart failure is a notoriously difficult task, even among experienced cardiologists. The diagnosis is usually done by the analysis of an echocardiogram by a cardiologist, who evaluates the ejection fraction of the heart, a metric that indicates how much blood is pushed through the heart’s chambers with each contraction. This examination requires specialized gear and trained personnel, leaving heart failure diagnosis under in-clinic expert supervision and a series of complex and costly clinical tests.

A COLLABORATION BETWEEN EXPERTS IN MEDICAL TECHNOLOGY AND ARTIFICIAL INTELLIGENCE

In order to help patients and cardiologists better deal with acute forms of heart failure, HeartKinectics, a Belgian-based Medtech company, is developing a non-invasive cardiac monitoring solution for patients to more easily monitor their heart for a regular follow-up and to improve their quality of life.

This team of experts is looking to streamline the process of heart failure diagnostics by using a wearable device and combining it with Artificial Intelligence in order to predict heart failures based on the signals of their wearable device, called KINOTM. KINO.md directly measures the cardiac mechanical activity of the heart. Data from KINO.md are sent to the cloud computing facility where Kinetic Energy (KE) and 100’s of parameters are analyzed to get a direct overview of a patient’s cardiac health; In addition, there is a web application for healthcare professionals that stores all patient’s measurements and issues a notification of suspicion of heart failure diagnostics from the reference measurements.

Kantify is a Belgian-based Machine Learning Solution development company, recognized for its excellence in Machine Learning and Life Sciences. In 2019, Kantify announced a world-first AI solution for predicting atrial fibrillation (arrhythmia) episodes at individual level, using patient’s RR-intervals data collected through a Holter monitor. Since then, Kantify has carried out numerous research cases of diagnosis and predictive analytics in life sciences, from automating analysis of signal data, to pathology detection from image data, and enhancing the drug discovery process by leveraging complex molecular databases.

DETECTING HEART FAILURE AT HOME

Kantify and HeartKinetics teamed up in a collaboration project to help patients suffering from cardiovascular diseases by developing an AI diagnosis tool to detect heart failure on an individual level, solely on patient’s signal data recorded through HeartKinetics’ KINO.md prototype medical device on the cardiac mechanical activity.

To test the feasibility and to validate the performance of an AI model, the collaboration initially started as a proof of concept study. Kantify developed an AI model to detect heart failure and classify it into four classes: no heart failure, and three sub-types of heart failure, namely low ejection fraction, mid-range ejection fraction, and preserved ejection fraction. The study included 500 records from 200 patients, presenting either no sign of heart failure, or confirmed observations of one of the 3 classes of heart failure.

GROUNDBREAKING RESULTS

In order to evaluate how well the AI model could distinguish both heart failure, the algorithm was trained on data originating from a subset of patients and tested on data from another set of patients that the algorithm had never seen, but for which clinical results were available, in order to compare these with the predicted results from the algorithm. Kantify trained two distinct algorithms - a “binary classification” algorithm, which predicts whether a patient wearing the Kino device is diagnosed with heart failure or not, and a second, “multi-class classification” algorithm, that attempts to predict whether a patient wearing the Kino device is at suspicion of heart failure, and if so, which type of heart failure they might be affected by.

Figure 1 General overview of the method: from data collection to patient classification (Kinetic Energy (KE), Power (Pow) are calculated for the Linear (Lin) and Rotational (Rot) dimensions.

The evaluation of the model showed very strong results in binary classification (presence or absence of heart failure) with an Area Under the Receiver Operator Curve (a technical metric where higher values indicate better performance) of 94% . Kantify’s AI algorithm managed to also achieve strong results in precise classification into 4 categories, with an Area Under the Receiver Operator Curve of 84%. Kantify & HeartKinetics, with this proved to be able to detect heart failure with preserved ejection fraction, a case that is very difficult to diagnose, even for cardiology experts in a clinical environment.

The study proved that HeartKinetics’ expertise in heart monitoring and signal processing combined with Kantify’s expertise in artificial intelligence has a great potential in helping patients and providing them with a tool to detect various cases of heart failures at any time. Timely diagnostics further opens the door for quick patient care by reducing the time between a heart failure develops and patient’s treatment at hospitals. This will help an earlier treatment and reducing risks of cardiac arrests or further complications. Furthermore, by better detecting when patients are not undergoing heart failure, the AI-powered Kino device can reduce the time it takes to correctly orient patients toward the correct specialty.

Dr Almorad, HeartKinetics CMO concludes ”The very promising results of this preliminary study will enable the development of a solution that will be widely available and easily used by patients and physicians. Heart failure patients and their caregivers will have a reliable way to monitor their heart health at their fingertips.”

Heartkinetics and Kantify are continuing their collaboration to bring the AI-powered Kino device to market in 2022.

Get in touch !