[ad_1]
A new model of artificial intelligence (AI) shows great accuracy for the diagnosis of pediatric conditions compared to the initial diagnosis by a medical examiner.
The model was also able to distinguish between less urgent common conditions and potentially life-threatening situations, according to a report from Huiying Liang, Guangzhou Women's and Children's Medical Center of the Guangzhou Medical University in China , and his colleagues.
The study, published online today in Nature Medicine, shows that this AI framework can mimic the clinical reasoning of human physicians and use machine learning to extract clinically relevant text from electronic health records (EHRs) in order to accurately predict the diagnosis of medical conditions. a patient, conclude Liang and his colleagues.
Rod Tarrago, MD, head of medical information and medical director of medication safety at Seattle Children's Hospital in Washington, said Medscape Medical News he sees this model as an exciting step forward. But, he added, this tool could help doctors rather than replace them.
It is encouraging to see such a model in pediatrics, which has fewer patients and has more difficulty feeding clinical trials, and which could benefit from automated learning models trained on large datasets. This model also differs from some previous models in that it relies on text rather than imaging to make diagnoses, he notes.
The system uses an automatic natural language processing and one of the authors, Kang Zhang, MD, PhD, said Medscape Medical News the template can be adapted to any language. Zhang is affiliated with the Guangzhou Women's and Children's Medical Center and the University of California San Diego.
The researchers compared the ability of the AI system and doctors to diagnose a list of conditions such as asthma, encephalitis, sinusitis and pneumonia. Physicians manually clbadified 11,926 EHRs into an independent cohort of pediatric patients.
Twenty physicians were grouped by experience and skill into two junior and three senior groups. One physician from each group read a subset of the clinical notes from the data and made a diagnosis for each patient.
The researchers attributed the performance to an F1 result, a precision and recall measure, which determined that the AI system outperformed the two older age groups, but that the score was slightly lower than the three groups of older physicians.
Specifically, the average F1 score for the AI model was 0.885. The F1 scores for the beginning physician groups averaged 0.841 and 0.839, and the F1 score for the experienced physician groups averaged 0.907, 0.915, and 0.923.
"This result suggests that this model of AI can potentially help young doctors to make a diagnosis but can not necessarily outperform experienced doctors," write the authors.
The authors also highlight the accuracy of the system in diagnosing hazardous conditions.
"Our system has been able to achieve this goal in several disease categories, as shown by its performance for acute exacerbations of asthma (0.97), bacterial meningitis (0.93) and for several diagnoses related to generalized systemic diseases, such as chickenpox (0.93), influenza (0.94), mononucleosis (0.90) and roséol (0.93), all of which can have potentially lethal, so that an accurate diagnosis is of utmost importance, "they write.
The authors see several uses of the framework in clinical practice. When patients enter an emergency department or an emergency care center, for example, the algorithm, which uses basic information, vital signs, and examination notes physical, can give priority to patients who need to see a doctor first. This could reduce waiting times and improve access to care.
Another application of the artificial intelligence system is to aid in the diagnosis of rare or complex diseases. The system was "trained" with 101.6 million data points from 1.4 million pediatric patient visits between January 2016 and July 2017 in a major referral center in Guangzhou. in China. The size of the dataset can help eliminate the bias of doctors who diagnose what they have seen from their own experience.
Tarrago said that diagnoses of the most complex and inconsistent conditions in the presentation would be the real test of the model.
"The possibilities are certainly there," he said, but added that the diagnoses presented in this study are not necessarily the areas in which human physicians have many difficulties.
The authors conclude that while this system may have the greatest impact in countries such as China where the ratio of health care providers per population is small, "the benefits of such a system will probably be universal".
"I think it's a big step in pediatrics and artificial intelligence," Tarrago said. "They have the benefit of having such a huge sample size.Now the question is, how does it work with different types of health care models?" That was more focused on ambulatory environments. "
Tarrago said that he would like to see models like this one eventually become a parallel partnership with the doctor and no longer a last resort when the doctor has trouble diagnosing it. He would like to see a real-time interaction where the doctor could "argue" with the AI system to make a diagnosis.
In such a scenario, he said, "Not only does he learn from us, but we continue to learn from it."
The study was funded by China's National Research and Development Program, the National Natural Science Foundation of China, the Guangzhou Children's and Women's Medical Center, the Guangzhou Regenerative Medicine and Health Laboratory. The authors of the study and Tarrago did not reveal any relevant financial relationship.
Nat Med. Posted online February 11, 2019. Summary
Follow Medscape on Facebook, Twitter, Instagram and YouTube
[ad_2]
Source link
