The Electronics and Telecommunications Research Institute (ETRI) has developed a non-face-to-face medical technology that uses artificial intelligence (AI) to aggregate medical intelligence from various institutions and predict a patient's health status with 90% accuracy, raising expectations for the establishment of a non-face-to-face collaborative medical system utilizing medical AI in the post-COVID era.

The Electronics and Telecommunications Research Institute (ETRI) has developed 'Dr. AI,' an artificial intelligence attending physician that integrates medical intelligence systems built across various hospitals to precisely analyze a patient's current condition and rationally predict their future health. It was announced on the 27th that it had developed 'AI'.

Doctor AI, developed by ETRI, assists in medical care by utilizing the EMR-based medical intelligence of each hospital simultaneously (ensemble) instead of integrating EMRs.

In other words, it has the effect of jointly utilizing medical data from other institutions without directly accessing sensitive information.

It amounts to indirectly turning medical information by institution into big data.

When a patient's current information is entered into Doctor AI, the medical intelligence of each institution analyzes it individually, integrates the results, adjusts for errors, and selects the optimal prediction.

It showed about 10% higher accuracy compared to using only medical intelligence from a single institution, which is because the predictions differ slightly depending on the patient group data according to hospital characteristics for each medical intelligence.

The research team improved accuracy by collaborating with medical intelligence trained on different data from each institution.

ETRI, together with Seoul Asan Medical Center, Ulsan National University Hospital, and Chungnam National University Hospital, utilized the EMRs of approximately 740,000 cardiovascular disease patients to secure a prediction accuracy of over 90%.

The core technologies of Doctor AI include ensemble medical intelligence (analysis of prediction trends and errors by institution), time-series EMR medical intelligence (analysis of prediction basis and health status), and multimodal medical intelligence (learning medical data).

So-called Ensemble Medical Intelligence identifies a patient's future health condition based on the most suitable medical data from hospitals nationwide equipped with Doctor AI, regardless of which hospital is visited.

For example, if chronic respiratory diseases diagnosed at local screening centers are analyzed and predicted using Doctor AI by leveraging the medical intelligence of large hospitals with sufficient accumulated cardiovascular data, more comprehensive and detailed analysis and prediction become possible.

It is even possible to predict that it is a disease that could cause serious heart damage in about two years..

The research team utilized time-series EMR medical intelligence to design different analysis weights and concentrations for factors such as hospital visit frequency and examination items, enabling more precise predictions.

Medical data used in time-series analysis requires a prediction method that considers the unique characteristics of EMRs, such as patients' irregular visit intervals and various types of tests, and this technology enabled us to improve accuracy.

Multimodal medical intelligence learns and utilizes not only EMR data but also cardiac CT image data, thereby improving the accuracy of cardiovascular disease prediction and assisting in personalized patient treatment.

ETRI plans to build medical intelligence in every hospital and improve accuracy by training deep learning on a vast number of cases that humans cannot handle.

It was stated that the purpose of the research is to ensure a healthy life for the public by actively utilizing even similar cases for prediction and diagnosing major diseases early.

Choi Jae-hoon, a principal researcher at ETRI and head of Doctor AI technology development, said, “Not only primary and secondary hospitals, where patient data is relatively scarce, but also large hospitals can achieve effects similar to collaborative treatment by simultaneously utilizing medical intelligence from hospitals with different patient populations. We expect that this will enable the standardization of medical standards to be raised.”

This study was conducted as part of the Ministry of Science and ICT's 'Development of AI Doctor Technology for Cardiovascular Diseases' project.

Dr. Seung-Hwan Kim of the Medical Information Research Division at ETRI served as the overall project manager, and Asan Medical Center and Daia Information Systems participated as joint research institutions, while Ajou University Hospital, Ulsan University Hospital, and Chungnam National University Hospital provided assistance through commissioned research.