AI is now able to diagnose rare eye diseases better than specialists

After the journalist, laborer, dermatologist, now the eye doctor will no longer be able to keep his position?

Diagnosing diseases is one of the practical benefits of machine learning system . Machines have information, lots of information, and they are capable of handling all this data in ways that the human brain cannot.

More importantly, machines can cure the "hardest" cases, the rare diseases that even the best doctors in the industry have neglected to neglect. AI system. However, machines are used to help people, not to replace them.

To accomplish this, a group of Chinese doctors and computer scientists introduced a machine learning algorithm to identify congenital cataract disease, a rare eye disease that causes 10% of children in the world completely lose sight. This algorithm has the ability to diagnose disease more than 90%, on par with the best eye doctor in the world. It is described in detail in the latest issue of Nature Niomedical Engineering.

Machines can cure the most "difficult" cases.

However, it should be noted that medicine is not only a science, but an art. The interaction between doctors and patients is difficult to measure its importance. A problem when treating with computer-made methods, it is wrong feeling, insecurity.

If the doctor made a mistake - maybe because he was negligent, but not everyone is like that - but when a machine fails, the error is that machine.

Even so, machines have a huge role."Machinery has the advantage of automation, objectivity and accuracy, but the ability to communicate and interact with people is indispensable in treatment" , co-author of the study, Haotian Lin , Professor of Ophthalmology at Sun Yat-sen University said.

"For doctors, technology is not enough to decide 100% to be sure what is the best therapy, and therefore they should take advantage of the suggestions coming from machines to identify and prevent school. incorporating false diagnosis based on their own knowledge, " he said. "The results of our analysis indicate that artificial intelligence and human intelligence have their own strengths and limitations. "

Failure to recognize or misdiagnose is common in patients with rare diseases, and this is true in developing countries, such as China.Cataract is a disease that makes it difficult for doctors to restore the initial state of the eye to be treated promptly as well as to apply the right treatments.

Patients with cataracts.

This algorithm is based on Convolutional Neural Networks (CNNs) technology , a type of machine learning that attempts to mimic the process of neurological processing occurring in the cortex of animals. CNNs are widely used for image recognition work, but are also applied to other areas, such as Go Go (Google Play Alpha Go flag player also uses it), speech analysis language and discover new pharmaceuticals.

The process involves loading images of how the disease for the AI ​​system, including cases of vitreous conditions, until it learns the presence of symptoms, then diagnoses disease success.

Researchers here have invented 3 useful networks for 3 different cataract identification jobs.The first network was used to screen patients from healthy people; The second is to identify patients with cataracts, and the third is to help eye doctors make treatment decisions. All three are integrated into a cloud - based platform called CC-Cruiser.

CC-Cruiser is a very promising machine, but it is not ready to be used right now.

The actual test of this software is based on 50 cases of disease selected by specialists and includes many "difficult to cure" diseases. The effectiveness of CC-Cruiser is compared with 3 different levels of doctors: practice doctors, doctors with clown license and expert doctors.

In normal cases, CC-Cruiser does not misdiagnose or make inaccurate conclusions. No level of human doctor achieved such results. Only when continuing to carry out the tasks of making decisions to apply therapies later does the new system succumb to more false diagnoses.

CC-Cruiser is a very promising machine, but it is not ready to be used right now."Currently, it has been used at three non-specialized hospitals to assess the CC-Cruiser medical applications , " Lin said. "However, because of respect and desire to protect human life, the medical field always keeps a cautious attitude when using new technology. It is also necessary to test it in many other reviews before they We can officially bring AI into medical help ".