April 10, 2021

Artificial intelligence can already diagnose childhood ailments with the precision of a pediatrician | Technology

Artificial intelligence can already diagnose childhood ailments with the precision of a pediatrician | Technology

The Impact of artificial intelligence on health it grows day after day, and the scientific community does not stop looking for new ways to exploit its full potential. This Monday, an international team has announced in an article the development of a machine capable of generating a wide range of diagnoses of childhood ailments. The system has been trained with the data of more than 500,000 patients treated in a reference healthcare institution located in Guangzhou (China). The researchers emphasize that this technology can help doctors to understand more quickly which patients need priority attention and to more accurately guess the diagnosis of rare or dangerous diseases.

The authors of the article, published in Nature Medicine, explain that the availability of medical information has grown exponentially in recent years, which complicates the decision making by doctors. The ability of artificial intelligence to analyze large amounts of data can help reduce this burden and facilitate the proper assessment of each clinical case, they say. But getting the machines to be able to correctly interpret the data is not an easy challenge to overcome, remember also.

The model they have developed is based on the exploitation of deep learning and natural language processing. "This allows us to extract key information freely distributed in patients' electronic health records," explains Kang Zhang, a researcher at the University of San Diego (California, USA) and a member of the team responsible for the project. After being trained with data manually recorded by pediatricians, the machine progressively acquires the ability to automatically incorporate and classify the relevant information and, subsequently, to make diagnoses.

Zhang and his colleagues say that the greater the amount of data on which the system is fed, the more its efficiency increases. In this work they have used data from almost 1.4 million pediatric appointments of more than half a million children and adolescents under age. "Our artificial intelligence system can imitate a human doctor and use all the health information to make a diagnosis," says Zhang.

"Our artificial intelligence system can imitate a human doctor and use all health information to make a diagnosis"

When comparing the assessments of the state of health generated by this artificial intelligence with those previously made by human pediatricians, the researchers found a similar level of precision between the two for a good number of childhood complaints (from the cold and the flu to diseases of the type). neurological). In some cases, the machine even became more accurate than less experienced doctors among those who had made the diagnoses.

Zhang says the system can potentially diagnose any type of pediatric disease, although it does not detract from what human professionals can contribute. "With more training, this system can get to perform most diagnoses with minimal supervision of doctors. But you can never completely replace a human ", ditch.

A mysterious technology

The expert of the Institute of Corpuscular Physics (CSIC and University of Valencia) Francisco Albiol highlights "the scope" of the study, which has involved "a large deployment of different profiles and workers to collect data, tag them, train the mathematical models, and check the results of part of the sample ". Ignacio Hernández, a doctor at the Ramón y Cajal Hospital in Madrid, believes that the most novel aspects are the breadth of diagnoses he is able to cover and the large amount of information he analyzes. "The previous models looked only very specific and structured data," he compares.

Hernández, also co-founder of the companies Savana Y MendelianOn the other hand, he believes that the study is "excessively opaque" in terms of describing how technology works. machine learning used. The expert states that in this area it is normal to give little details about the mechanism of developed systems, not so much because of secrecy but because it is often difficult for the developers themselves to identify "what variables the machine looks at and why when it operates". However, in his opinion, in this case the authors "go further". "They do not explain anything at all about their model. It is disturbing and in a way even wrong, "he says.

The doctor also believes that studies like this have the limitation of showing results only "on paper" and not in a real plan. "The clinical practice is very contextual, it depends on aspects such as the socio-health and economic conditions of the patient and many other elements," he argues. "If we do not have a live demonstration, in perspective, we can not yet say that such a system solves problems."

José Luis Salmerón.
José Luis Salmerón.

A paradigm shift

José Luis Salmerón, an expert in artificial intelligence at the Pablo Olavide University in Seville, does not see "surprising" information with respect to previous works, but he does consider it promising that "from the health fields they begin to be aware of what specialists in this field can contribute in that sector. " Hernández agrees with this argument. "By having these tools, we have to redraw in our head what a health system is like. People are going to have access to them on-line, on your mobile. Many elements that we take for granted will change, "he says.

Zhang and his colleagues believe that the proposed model will accelerate the diagnosis of the most common diseases and will give physicians the possibility to focus on patients who need urgent care. In this way, they underline, it will be possible to optimize health care and reduce waiting times.

The system can also help doctors take into consideration more hypotheses of assessment than they can formulate based on their own experience when faced with rare or complex ailments, they add. The potential benefits, according to these scientists, will be particularly beneficial in areas of the world with limited health resources, such as rural areas in China.


Source link