قالب وردپرس درنا توس
Home https://server7.kproxy.com/servlet/redirect.srv/sruj/smyrwpoii/p2/ Health https://server7.kproxy.com/servlet/redirect.srv/sruj/smyrwpoii/p2/ AI is good (maybe too good) to predict who will die prematurely

AI is good (maybe too good) to predict who will die prematurely



  Is AI good (maybe too good) when predicting who will die premature

Can AI predict when you die?

Credit: Shutterstock

Medical researchers have unlocked restless AI capability: predicting early human death

Scientists recently trained an artificial intelligence system to evaluate decades of general health data provided by more than half a million people in the United Kingdom. They then instructed AI to predict whether individuals are at risk of premature death ̵

1; in other words, earlier than life expectancy – from a chronic disease, they said in a new study.

of the AI's algorithms were "significantly more accurate" than the predictions provided by a model that did not use machine training, the author of a leading study, Dr. Stephen Wang, an assistant professor of epidemiology and informatics at the University of Nottingham ) in the UK said in a statement. [Can Machines Be Creative? Meet 9 AI ‘Artists’]

In order to assess the probability of premature mortality of the participants, researchers try two types of AI: "Deep Learning", in which layered information processing nets help the computer learn from examples; and "casual forest," a simpler type AI that combines multiple tree-like models to look at the possible results.

They then compare the conclusions of the AI ​​models with the results of a standard algorithm, known as the Coke model. 19659005] Using these three models, scientists evaluated the data in the British Biobank, an open access database of genetic, physical and health data, presented by more than 500,000 people between 2006 and 2016. During that time, almost 14,500 of the participants died, from cancer, heart disease and respiratory diseases. Different variables

All three models found that factors such as age, gender, smoking history, and previous cancer diagnoses are the main variables for assessing the likelihood of early death of a person. But patterns differ from other key factors, researchers found.

The Coke model relies heavily on ethnicity and physical activity, whereas machine learning models are not. By way of comparison, the incidental forest pattern places more emphasis on the percentage of body fat, the waist circumference, the amount of fruits and vegetables people eat, and the tone of the skin, according to the study. For the deep learning model, the most important factors include exposure to risks related to work and air pollution, alcohol consumption and the use of certain medicines.

When the full crunching number is made, the deep learning algorithm has provided the most accurate predictions to properly identify 76% of those who died during the study period. By way of comparison, the random forest model correctly predicts about 64% of premature deaths, while Coke's model identifies only about 44%.

This is not the first time that experts have used the predictive power of AI for healthcare. In 2017, a different team of researchers demonstrated that AI may learn to detect early signs of Alzheimer's disease; their algorithm has evaluated the brain scan to predict whether a person could develop Alzheimer's disease and did it with about 84% accuracy, Live Science reported.

Another study found that AI could predict the onset of autism within a 6-month period. babies who have been at high risk of developing the disease. One more study may detect signs of onset diabetes by retinal scan analysis; and another – using retinal scan data – predicts the likelihood of a heart attack or stroke.

In the new study, scientists have demonstrated that machine training – "with careful tuning" – can be used to successfully predict death results over time, the co-author of the study, Joe Kai, a professor of primary medical care at the UN, 19659005] During the use of AI, this way may be unknown to many health professionals by presenting the methods used in the study

The results were published today (March 27th) in the journal PLOS ONE.

Originally posted on Science .


Source link