[ad_1]
Millions of dollars is used to make artificial intelligence programs that read x-rays and other therapies are expected to be able to see what doctors are looking at but sometimes miss out, such as lung cancer. New research says this and algorithms they can also see something that doctors do not look at in a mirror like this: the type of patient.
The researchers and other AI medical experts say the results make it even more important than ever to see if medical algorithms work fairly for different people. Coping with this work: The authors do not know what makes people think they are.
Evidence that algorithms were able to calculate color from human medical imaging was based on five-dimensional imagery used in radiology research, including chest and hand x-rays and mammograms. The images included patients identified as black, white, and Asian. For each type of analysis, the researchers taught the algorithms using the written images and the self-explanatory competition. He then challenged algorithms to predict the type of patients in a variety of images, without letters.
Radiation specialists do not consider a person’s color – which is not natural – to be visible on images that look under the skin. However some algorithms somehow proved that they could be found correctly for all three types of colors, as well as for different body types.
In most options, algorithms can detect which of the two images was from Black Man more than 90 percent of the time. Even the worst algorithms managed to do 80 percent of the time; the best was 99 percent correct. The program of results and connection code was posted online late last month by a team of more than 20 researchers with medical expertise and machine learning, but this study has not yet been peer-reviewed.
The results have raised new concerns that the AI program could exacerbate medical inequalities, with studies showing that black patients and other marginalized groups often receive less attention than the rich or white.
Machine learning algorithms are developed to read medical images by feeding multiple samples of tumors. By digging multiple samples, algorithms are able to study the pixels that are read by the letters, such as the shape or shape of the lungs. Some algorithms make it easier for veterans to diagnose cancer or skin problems; there is evidence that he is able to recognize the symptoms of the disease invisible to human experts.
Judy Gichoya, a radiologist and assistant professor at Emory University who participated in the study, says the revelation that graphical algorithms can “see” the internal competition makes them also learn to associate inappropriately.
Many of the therapies used in the training of algorithms often have racial differences in disease and medical care, due to their historical and economic background. This could lead to a systematic analysis of the variables to apply its concept of patient type as a shortcut, suggesting that the disease-related genetic predisposition is based on its studies, not the medical complications that radiologists look for. Such a procedure can give some patients a false or false diagnosis. The algorithm can interpret the differences between black and white people with similar symptoms of the disease.
“We need to educate people about the problem and find out what we can do to solve it,” Gichoya said. Participants came from organizations such as Purdue, MIT, Beth Israel Deaconess Medical Center, National Tsing Hua University in Taiwan, University of Toronto, and Stanford.
Previous research has shown that clinical algorithms have caused conflict in the delivery of care, and that visual algorithms can perform differently in different groups. In 2019, a widely used approach to improving patient care became available lack of black people. In 2020, researchers at the University of Toronto and MIT showed that algorithms trained to describe such things as pneumonia x-ray sometimes react differently to people of the same sex, age, race, and type of medical insurance.
[ad_2]
Source link
