At the 22nd China Cerebrovascular Disease Forum held recently, a unique competition - AI big model PK human doctors has attracted attention.
This competition set up three distinctive cerebrovascular disease diagnosis and treatment cases. Human doctors and three AI models each analyzed the cases within a limited time and gave diagnostic results and treatment plans.
The AI models participating in the competition are 39AI, Doubao and DeepSeek (in-depth search).
Dr. Zhang Can, Department of Neurosurgery at Xuanwu Hospital, Capital Medical University, introduced that the 39AI big model is the first vertical medical big model registered by the National Cyberspace Administration of China and has received training on the consensus of more than 30 million medical literature around the world. The bean bread model has been trained in multiple corpus and has performed outstandingly in health science popularization, consultation and disease diagnosis and analysis. DeepSeek is a general-purpose large language model. The reasoning ability of its R1 model has been widely verified by various industries. This competition can test the differences between it and the model after secondary development of medical care.
The on-site judges scored the participating human doctors and AI models from five aspects: diagnosis accuracy, treatment plan accuracy, knowledge breadth, logical clarity, and answering efficiency.
The first clinical case is complicated. The case is a patient with a huge intracranial aneurysm. After previous surgery, it is not completely cured and has symptoms such as headaches. A new treatment plan needs to be designed.
What amazes the audience and human doctors that the AI model gives diagnostic results and treatment plans in just a few minutes and is displayed on the large screen in real time.
"By learning from a large number of literature, the AI model has paid attention to many details in the diagnosis, such as hypernatremia, hematoma on the lower limb puncture point, etc., and the differential diagnosis is also relatively comprehensive." Shi Huaizhang, chief physician of the Department of Neurosurgery of the First Affiliated Hospital of Harbin Medical University, said in a comment.
However, experts believe that although AI can provide rich information, it lacks the intuition and targeting of doctors. "Faced with complex cases, AI has given many choices, but failed to accurately judge the most suitable surgical plan for patients - high-flow bypass surgery, like experienced doctors," said Shi Huaizhang.
After the competition, human players and AI models won honors respectively.
Tian Qi, a doctor at the First Affiliated Hospital of Hebei Northern College, won the Guardian of Human Wisdom Award, Wang Yanan, a doctor at the Second Central Hospital of Baoding City, Hebei Province, won the Best Clinical Performance Award, and Zhang Yiming, a doctor at the You'anmen Hospital in Beijing, won the Best Potential Rookie Award. The Doubao big model won the Precision Diagnosis Award, and the 39AI model won the Best Treatment Solution Award. Although DeepSeek has not undergone intensive training for medical knowledge, it still has good performance and won the Most Innovative Potential Award.
Tian Qi said in an interview with a reporter from Science and Technology Daily that AI has obvious advantages in speed and comprehensive diagnosis, but human doctors are better in terms of targeted treatment and empathy.
This view resonates. Wang Yanan also said: "Doctors can communicate with patients and provide emotional support, and then develop more humane and personalized treatment plans."
Regarding this competition, Zhang Hongqi, director of the Department of Neurosurgery at Xuanwu Hospital of Capital Medical University, said: "AI answers quickly and has a wide range of details. It performs well in the diagnosis of routine diseases. Its accuracy is similar to that of humans and has better comprehensiveness. However, when facing emergencies, individualized treatments and scenarios that require humanistic care, AI has obvious shortcomings."
"Medical safety has always been the top priority, and speed is not a key factor. AI should be an auxiliary in medical care. Human doctors will still play a leading role in medical decision-making with their rich clinical experience, personalized treatment ability and humanistic care spirit." Wang Yanan said.
This competition provides new ideas for the cultivation of medical talents. "In terms of talent training in the medical industry, the development of diagnosis, treatment, Q&A, and examination assistants with the help of big models can help medical students check for deficiencies and cultivate comprehensive clinical thinking." Zhang Can said, but how to allow medical students to retain the unique advantages of human doctors while using AI technology is still a question that educators need to think deeply about.
Zhang Hongqi emphasized that AI and human doctors are not competitive, but partners that complement each other and develop in a coordinated manner. (Reporter Dai Xiaopei)
[Editor in charge: Ran Xiaoning]
Comment