Master clinical medical knowledge at certificated-doctor-level with deep learning model

Area of Science:

• Artificial Intelligence in Medicine
• Medical Education Technology

Background:

• Medical knowledge acquisition is a lengthy, intensive human process.
• Deep learning (DL) shows promise for addressing complex medical challenges.

Purpose of the Study:

• To develop a DL framework (Med3R) capable of mastering clinical medical knowledge at a certified doctor level.
• To evaluate Med3R's performance on a national medical licensing examination and in clinical diagnosis.

Main Methods:

• Med3R framework utilizes a human-like learning and reasoning process.
• The system was tested on the 2017 National Medical Licensing Examination in China.
• Med3R was applied to provide aided clinical diagnosis using real electronic medical records.

Main Results:

• Med3R achieved a score of 456 on the exam, surpassing 96.3% of human examinees.
• The AI system demonstrated more accurate and consistent clinical diagnosis results compared to human experts and baselines.
• Med3R successfully passed the written test of the National Medical Licensing Examination in China.

Conclusions:

• Med3R represents a significant advancement in AI for mastering medical knowledge.
• The framework shows potential for improving the accuracy and consistency of clinical diagnosis.
• Med3R could help alleviate the shortage of qualified doctors by providing computer-aided medical care.