A new study from Yonsei University and Yongin Severance Hospital has revealed that artificial intelligence models demonstrate a higher degree of accuracy in clinical diagnosis than human physicians. The research, published this week, showed the latest AI reasoning model, ‘o1’, achieved 94.3% accuracy in analyzing patient cases, surpassing the 85% average accuracy of medical professionals.
The study, which analyzed 1,426 patient cases from the medical education platform ‘Medscape’ spanning 2011 to 2024, included a comprehensive range of medical data – from patient histories and lab results to X-rays, CT scans, MRIs, and electrocardiograms. Researchers deliberately mirrored real-world clinical conditions to ensure the validity of the comparison. Earlier AI models, like GPT-4o, also outperformed physicians with 88.4% accuracy.
Although the findings suggest a potential for AI to surpass human capabilities in diagnostic precision, researchers emphasize that the technology is best suited as a support tool for doctors, rather than a replacement. “If AI is better at diagnosing than doctors, why do we need doctors?” is a question often posed, but the response, according to those in the field, is whether AI can hold a patient’s hand.
The increasing adoption of AI in healthcare is driven, in part, by the severe shortage of medical personnel in South Korea. According to a report from Bupyeong Serim Hospital, the country’s physician-to-patient ratio is significantly higher than in other OECD nations, placing a substantial burden on medical staff. AI-based solutions are being implemented in intensive care units to predict patient deterioration – including sepsis, cardiac arrest, and even mortality – allowing for faster and more informed decision-making. The technology acts as a “safety net” for medical professionals, according to Dr. Hwang Jun-ha of Bupyeong Serim Hospital.
Beyond diagnostics, AI is being utilized in a variety of hospital settings. Many institutions are now employing AI to automatically analyze chest X-rays and brain MRIs, identify anomalies, and even suggest treatment options for cancer patients based on genetic analysis and medical records. Large hospitals are actively integrating AI into inpatient care management, particularly in critical areas where rapid assessment is crucial. Forbes Korea reports that AI is evolving into a “clinical decision support infrastructure,” providing medical staff with critical warning signs and additional information.
Domestic hospitals have already begun implementing AI solutions, including platforms designed to expand AI capabilities across multiple diseases and undergo validation for international use. The development of Large Language Model (LLM) based ‘docents’ is also improving patient experience by providing guidance and enhancing accessibility. Analysis of Electronic Medical Records (EMR) data, powered by AI, is contributing to improvements in patient stability and predicting outcomes.
The Seoul Metropolitan Government recently reported that AI’s diagnostic accuracy reached 94% compared to 85% for medical professionals, but cautioned that the technology should be used as a tool to assist, not replace, doctors. The Ministry of Health and Welfare has not yet issued a formal statement regarding the implications of these findings for healthcare policy.
