Heart health at your fingertips

32% die of heart disease. 80% would be preventable

HeartShield develops solutions for recognizing, tracking and preventing heart disease using artificial intelligence, supported by scientific research.

Sign up for beta release

Email:

Risk detection accuracy results


We are constantly improving our predictive accuracy using new data
(Please sign up and measure anonymously to help increase our accuracy and help us save lives!)

The results below significantly outperform the state of the art in home diagnostics, as well as most widely used clinical risk profiling scores, without needing expensive devices

training set size

million heart beats

sensitivity

     

%

risk cases recognized

specificity

     

%

controls recognized


* Sensitivity: correctly identified percentage of individuals with risk factors. Specificity: correctly identified percentage of individuals without risk factors (controls).
Here, we define the "risk" of future heart disease the same way as the widely used Framingham Risk Score

As seen in

Supported by / awards

How it works

How it works

For many people, there are preventable causes of heart disease risk. For example, cholesterol can combine with other substances, and build up plaque in the arteries. Growing plaque can start blocking blood flow, and can even rupture, and cause a heart attack or stroke. These risks can be reduced with a healthier lifestyle, or with medical treatment, if they are recognized in time. Unfortunately, 16-78% are unaware of at least one risk factor according to the WHO.

HeartShield can recognize risk factors, and narrowed arteries, in time for early risk reduction, helping prevent dangerous disease events such as heart attacks in time - anywhere and anytime, for free, and in less than two minutes, without waiting times or syringes.

Why HeartShield?

We have developed a novel, patent pending artificial intelligence platform able to learn the signs of heart disease risk from patient data. Instead of using 19th/20th century statistics, as in clinical heart rate variability measurements, we allow computers to learn warnings signs autonomously in a data driven fashion. Crucially, the same artificial intelligence can learn from, and recognize risk from, most devices that can measure heart rate - including smartphones, smart watches, and clinical or portable ECGs.

Patent-pending AI

Able to learn about risk factors and pathologies through a novel combination of a physiological model of biological pacemaker neurons, and cutting-edge deep learning

Powered by big data

HeartShield has accumulated over 250 million heart beats to teach the AI, constantly adding more data and increasing its accuracy

Device-agnostic

The HeartShield AI works on smartphones, smart watches, tablets, laptops, clinical ECGs or oximeters - anything with optical or electrical sensors and sufficient frame rate

Better than the cutting edge

Peer-reviewed scientific papers show that HeartShield outperforms the best heart rate variability predictors in detecting heart disease, and is more reliable than blood test based risk scores (e.g. FRS) in recognizing coronary artery disease.

The Science

We have compared the algorithms behind HeartShield to state of the art approaches on several patient cohorts in multiple peer-reviewed scientific papers

NIPS (Neural Information Processing Systems) 2016 - Machine Learning for Healthcare (accepted)

Deep neural heart rate variability analysis (NIPS ML4HC 2016)

Abstract. Despite of the pain and limited accuracy of blood tests for early recognition of cardiovascular disease, they dominate risk screening and triage. On the other hand, heart rate variability is non-invasive and cheap, but not considered accurate enough for clinical practice. Here, we tackle heart beat interval based classification with deep learning. We introduce an end to end differentiable hybrid architecture, consisting of a layer of biological neuron models of cardiac dynamics (modified FitzHugh Nagumo neurons) and several layers of a standard feed-forward neural network. The proposed model is evaluated on ECGs from 474 stable at-risk (coronary artery disease) patients, and 1172 chest pain patients of an emergency department. We show that it can significantly outperform models based on traditional heart rate variability predictors, as well as approaching or in some cases outperforming clinical blood tests, based only on 60 seconds of inter-beat intervals.

CinC (Computing in Cardiology) 2016 (accepted)

Network analysis of heart beat intervals using horizontal visibility graphs

Abstract.Heart beat interval time series contain information predictive of heart disease, but most current predictors do not provide sufficient reliability for clinical use. Using several predictors improves predictive power, but the limit is not yet known, suggesting that not all the information in interbeat interval series has been captured by previous work. We convert heart beat time series into scale-free networks using horizontal visibility graphs (HVGs), which are well-suited to distinguishing deterministic dynamical systems from stochastic systems, allowing them to model new aspects of autonomic heart rate modulation. Based on the HVG, we introduce and evaluate a general class of predictors, which can be used to augment existing features used in heart rate variability (HRV) analysis, and which exhibit high predictive power for several types of heart disease. We show the statistical significance of these network predictors, and their competitive performance to popular statistical, geometric and non-linear features, on ICU and Holter ECGs, including several heart disease etiologies

IJCNN (International Joint Conference on Neural Networks) 2017 (accepted)

Deep Graph Embeddings for the Analysis of Short Heartbeat Interval Time Series Sudden cardiac death (SCD) constitutes a large proportion of cardiovascular mortalities, provides little advance warning, and the risk is difficult to recognize based on ubiquitous, low cost medical equipment such as the standard, 12-lead, ten second ECG. Autonomic abnormalities have been shown to be strongly predictive of SCD risk; yet current methods are not trivially applicable to the brevity and low temporal and electrical resolution of standard ECGs. Here, we build horizontal visibility graph representations of very short inter-beat interval time series, and perform unsuper- vised representation learning in order to convert these variable size objects into fixed-length vectors preserving similarity rela- tions. We show that such representations facilitate classification into healthy vs. at-risk patients on two different datasets, the Mul- tiparameter Intelligent Monitoring in Intensive Care II and the PhysioNet Sudden Cardiac Death Holter Database. Our results suggest that graph representation learning of heartbeat interval time series facilitates robust classification even in sequences as short as ten seconds.

HeartShield Labs

Apart from the peer-reviewed and validated heart disease risk detection AI, we are also developing experimental technology for physiological time series analysis, including

Interested in trying our technology? Get in touch!

contact