The American Startup Cardiogram used data on people's heartbeat, obtained with the help of smart watches, to train Deepheart, a program that predicts the presence of various diseases.
The algorithm, based on the LSTM neuronal network, can predict diabetes with an 85% precision and the apnea with an 83% accuracy.
The prepublication of the article is available in Arxiv, and the results of the investigation were presented at the AAAI-2018 artificial intelligence conference.
Cardiogram developers with researchers at the University of California in San Francisco, under the direction of Mark Pletcher, used data from 14,000 participants in the study, who used a fitness application in Apple Watch.Participants also provided information about their medical conditions.
The researchers were interested in diabetes, high cholesterol, hypertension and sleep apnea.Part of the data was used for training of the LSTM neuronal network without a teacher.Thanks to this, the algorithm has learned to recognize variations in the human heart.
Data on differences in heartbeat were then used in the test and the algorithm could successfully identify diabetes (84.5%), high cholesterol (74.4%), hypertension (80.8%) and thesleep apnea (82.9%).
Although the use of Deepheart has proven effective, the authors point out that in the future, it is necessary to take into account a large number of additional variables: the age and sex of the participants, smoking, alcohol consumption,etc
In general, scientists believe that this method can be used to prevent the appearance of states studied through physical activity trackers and smart watches from different manufacturers.
Artificial intelligence can also map the brain and detect Alzheimer with 10 years in advance.
Elizabeth Ivtushok
Text translated by María Cervantes
This news has been originally published in N+1, technology that adds.
About N+1: It is the first online magazine of scientific and technological dissemination that allows the total or partial reproduction of its contents by the media, bloggers and influencers, making the mention of the text and the link to the web: “This news hasbeen originally published in the magazine N+1, technology that adds ”.
