A team of scientists at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a machine learning AI system that is capable of assessing individual risk of heart disease by analyzing your heart’s overall electrical activity.
“RiskCardio”, as it is named, uses data obtained from a patient’s 15-minute reading of his or her electrocardiogram (ECG) signal. The final analysis produces a sort of score, which is then categorized into different heart-based risk factors for the owner of the data.
The training data used to establish RiskCardio’s base performance parameters came from the ECG readings for the test patients. Each reading was separated into individual, consecutive heartbeats, which were then classified into sections, and are then assessed for possible health risks. After using countless sample data from these sections, the AI is then trained to identify which of these beats most closely resemble the sample data of those who actually died from their predicted heart failures.
Indeed, when compared to similar types of assessment AI, this new ECG-based heart analysis AI directly assessed that survivors of Acute Coronary Syndrome (ACS) were at least seven times more likely to die of heart disease, as opposed to the “three times more likely” assessment of older AI models.
There is also the deliberate move to exclude other types of traditional data from the score reading, such as age or weight, simply completing the assessment with the ECG reading alone. This is most likely to provide a separate comparison that would allow medical professionals to confirm the risk level from two or more independent readings.
The researchers admit that the technology is far from perfect as it is. According to the official press release, “the signals are very long, and as the number of inputs to a model increase, it becomes harder to learn the relationship between those inputs.”
Nevertheless, after this milestone, the research team is planning to expand the AI’s dataset further, to include a wider variety of patients of different ages, ethnicities, and to both biological genders. More importantly, perhaps, they want to test their AI model on heart disease data that was previously assessed incorrectly, in hopes of making the AI even more robust.