Artificial intelligence in medicine may be able to distinguish between races

According to a study of Massachusetts Institute of Technology (MIT), the iartificial intelligence (AI), can predict the races of patients through their medical images. Ziad Obermeyer, associate professor at the University of California at Berkeley, assured that the fact that the algorithms see the race It could be dangerous.

The research was carried out using a set of public and private data. Among them are chest x-rays, extremity x-rays, chest CT scans, and digital mammograms.

Apart from the above, the team trained a learning model to identify white, black or asian race despite not having explicit mention in the images of the patient. To achieve this, a series of experiments with artificial intelligence were carried out to investigate the possible mechanisms of race detection.

Therefore, variables such as differences in anatomy, bone density, imaging resolution, among other. The study highlighted that despite variabilities, artificial intelligence continued to detect race from chest X-rays.

Leo Anthony Celi, a researcher at MIT and an associate professor of medicine at Harvard Medical School, commented that algorithms can amplify existing disparity and inequality. Therefore, he considered it important to reflect and reconsider whether humanity is prepared to bring AI to the patient’s bedside.

The study, called “AI recognition of patient race in medical imaging: a modeling study,” was published in Lancet Digital Health on May 11, 2022. Celi and Marzyeh Ghassemi wrote the article along with 20 other authors from four countries.

Scientists first showed that artificial intelligence was capable of predicting races in different imaging modalities. Likewise, such a mechanism guess various data sets, clinical tasks and academic centers.

See also  Argentina: Advances in the regulation on medicinal use for cannabis-based products and derivatives.

For this, three large datasets of chest radiographs and model was tested on an unseen subset of data used. Subsequently, they trained the racial identity detection models for non-thoracic X-ray images of multiple body locations. This in order to see if the performance of the model was limited to chest radiographs.

In order to explain the behavior of the model, the team covered several bases. Among them, differences in physical characteristics between different racial groups, disease distribution, and site- or tissue-specific differences. Likewise, the effects of social prejudices, environmental stress within the study, the ability of artificial intelligence to detect races and whether the regions of the image contributed to recognizing these were covered.

In this way, the scientists discovered that the ability of artificial intelligence to predict races through diagnostic labels was lower. For their part, the models based on chest X-ray images had better prediction.

However, the scientists recognized that the availability of racial identity labels is limited. As a consequence of the scenario described above, they focused on Asian, black, and white populations.

Another work by Ghassemi and Celi, directed by the MIT student, Hammaad Adamdiscovered that the artificial intelligence can also identify the races declared by the patient himself. This from the clinical notes, even when these notes lack explicit indicators of race.




Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Social Media

Most Popular

On Key

Related Posts