User login
Imagine that instead of a patient visiting their doctor for blood tests, they could rely on a noninvasive at-home test to predict their risk of diabetes, a disease that affects nearly 15% of U.S. adults (23% of whom are undiagnosed), according to the U.S. Centers for Disease Control and Prevention.
This technology could become a reality thanks to a research team that developed a machine learning algorithm to predict whether people had type 2 diabetes, prediabetes, or no diabetes. In an article published in BMJ Innovations, the researchers describe how their algorithm sorted people into these three categories with 97% accuracy on the basis of measurements of the heart’s electrical activity, determined from an electrocardiogram.
To develop and train their machine learning model – a type of artificial intelligence (AI) that keeps getting smarter over time – researchers used ECG measurements from 1,262 people in Central India. The study participants were part of the Sindhi population, an ethnic group that has been shown in past studies to be at elevated risk for type 2 diabetes.
Why ECG data? Because “cardiovascular abnormalities and diabetes, they go hand in hand,” says study author Manju Mamtani, MD, general manager of M&H Research, San Antonio, and treasurer of the Lata Medical Research Foundation. Subtle cardiovascular changes can occur even early in the development of diabetes.
“ECG has the power to detect these fluctuations, at least in theory, but those fluctuations are so tiny that many times we as humans looking at that might miss it,” says study author Hemant Kulkarni, MD, chief executive officer of M&H Research and president of the Lata Medical Research Foundation. “But the AI, which is powered to detect such specific fluctuations or subtle features, we hypothesized for the study that the AI algorithm might be able to pick those things up. And it did.”
Although this isn’t the first AI algorithm developed to predict diabetes risk, it outperforms previous models, the researchers say.
The team hopes to test and validate the algorithm in a variety of populations so that it can eventually be developed into an accessible, user-friendly technology. They envision that someday their algorithm could be used in smartwatches or other smart devices and could be integrated into telehealth so that people could be screened for diabetes even if they weren’t able to travel to a health care facility for blood testing.
The team is also studying other noninvasive methods of early disease detection and predictive models for adverse outcomes using AI.
“The fact that these algorithms are able to pick up the things of interest and learn on their own and keep learning in the future also adds excitement to their use in these settings,” says Dr. Kulkarni.
A version of this article first appeared on Medscape.com.
Imagine that instead of a patient visiting their doctor for blood tests, they could rely on a noninvasive at-home test to predict their risk of diabetes, a disease that affects nearly 15% of U.S. adults (23% of whom are undiagnosed), according to the U.S. Centers for Disease Control and Prevention.
This technology could become a reality thanks to a research team that developed a machine learning algorithm to predict whether people had type 2 diabetes, prediabetes, or no diabetes. In an article published in BMJ Innovations, the researchers describe how their algorithm sorted people into these three categories with 97% accuracy on the basis of measurements of the heart’s electrical activity, determined from an electrocardiogram.
To develop and train their machine learning model – a type of artificial intelligence (AI) that keeps getting smarter over time – researchers used ECG measurements from 1,262 people in Central India. The study participants were part of the Sindhi population, an ethnic group that has been shown in past studies to be at elevated risk for type 2 diabetes.
Why ECG data? Because “cardiovascular abnormalities and diabetes, they go hand in hand,” says study author Manju Mamtani, MD, general manager of M&H Research, San Antonio, and treasurer of the Lata Medical Research Foundation. Subtle cardiovascular changes can occur even early in the development of diabetes.
“ECG has the power to detect these fluctuations, at least in theory, but those fluctuations are so tiny that many times we as humans looking at that might miss it,” says study author Hemant Kulkarni, MD, chief executive officer of M&H Research and president of the Lata Medical Research Foundation. “But the AI, which is powered to detect such specific fluctuations or subtle features, we hypothesized for the study that the AI algorithm might be able to pick those things up. And it did.”
Although this isn’t the first AI algorithm developed to predict diabetes risk, it outperforms previous models, the researchers say.
The team hopes to test and validate the algorithm in a variety of populations so that it can eventually be developed into an accessible, user-friendly technology. They envision that someday their algorithm could be used in smartwatches or other smart devices and could be integrated into telehealth so that people could be screened for diabetes even if they weren’t able to travel to a health care facility for blood testing.
The team is also studying other noninvasive methods of early disease detection and predictive models for adverse outcomes using AI.
“The fact that these algorithms are able to pick up the things of interest and learn on their own and keep learning in the future also adds excitement to their use in these settings,” says Dr. Kulkarni.
A version of this article first appeared on Medscape.com.
Imagine that instead of a patient visiting their doctor for blood tests, they could rely on a noninvasive at-home test to predict their risk of diabetes, a disease that affects nearly 15% of U.S. adults (23% of whom are undiagnosed), according to the U.S. Centers for Disease Control and Prevention.
This technology could become a reality thanks to a research team that developed a machine learning algorithm to predict whether people had type 2 diabetes, prediabetes, or no diabetes. In an article published in BMJ Innovations, the researchers describe how their algorithm sorted people into these three categories with 97% accuracy on the basis of measurements of the heart’s electrical activity, determined from an electrocardiogram.
To develop and train their machine learning model – a type of artificial intelligence (AI) that keeps getting smarter over time – researchers used ECG measurements from 1,262 people in Central India. The study participants were part of the Sindhi population, an ethnic group that has been shown in past studies to be at elevated risk for type 2 diabetes.
Why ECG data? Because “cardiovascular abnormalities and diabetes, they go hand in hand,” says study author Manju Mamtani, MD, general manager of M&H Research, San Antonio, and treasurer of the Lata Medical Research Foundation. Subtle cardiovascular changes can occur even early in the development of diabetes.
“ECG has the power to detect these fluctuations, at least in theory, but those fluctuations are so tiny that many times we as humans looking at that might miss it,” says study author Hemant Kulkarni, MD, chief executive officer of M&H Research and president of the Lata Medical Research Foundation. “But the AI, which is powered to detect such specific fluctuations or subtle features, we hypothesized for the study that the AI algorithm might be able to pick those things up. And it did.”
Although this isn’t the first AI algorithm developed to predict diabetes risk, it outperforms previous models, the researchers say.
The team hopes to test and validate the algorithm in a variety of populations so that it can eventually be developed into an accessible, user-friendly technology. They envision that someday their algorithm could be used in smartwatches or other smart devices and could be integrated into telehealth so that people could be screened for diabetes even if they weren’t able to travel to a health care facility for blood testing.
The team is also studying other noninvasive methods of early disease detection and predictive models for adverse outcomes using AI.
“The fact that these algorithms are able to pick up the things of interest and learn on their own and keep learning in the future also adds excitement to their use in these settings,” says Dr. Kulkarni.
A version of this article first appeared on Medscape.com.