Application of DeepSeek-assisted problem-based learning in hematology residency training

Area of Science:

• Medical Education
• Artificial Intelligence in Medicine
• Hematology Training

Background:

• Traditional problem-based learning (PBL) is a cornerstone of medical education.
• Integrating advanced AI tools like large language models (LLMs) presents new opportunities to enhance training.
• Evaluating the efficacy of specific LLMs in specialized medical curricula is crucial.

Purpose of the Study:

• To assess the impact of integrating the open-source LLM DeepSeek into a hematology residency PBL curriculum.
• To compare the effectiveness of DeepSeek-assisted PBL with traditional PBL methods.
• To evaluate resident performance across various competency domains.

Main Methods:

• A non-randomized controlled trial involving 60 hematology residents divided into two groups.
• One group received traditional PBL; the other received DeepSeek-assisted PBL (using DeepSeek V3 and R1 models with AI-facilitated search).
• Learning outcomes were measured via post-course surveys and five standardized examinations.

Main Results:

• DeepSeek-assisted PBL showed significant advantages in case analysis, feedback quality, and clinical reasoning.
• The DeepSeek group outperformed in four of five examination domains, indicating improved academic performance.
• No significant difference was found in clinical skills, and participants raised concerns about AI accuracy.

Conclusions:

• DeepSeek integration can enhance clinical competence, diagnostic reasoning, and engagement in hematology training.
• Open-source LLMs offer scalable, cost-effective tools for augmenting medical education.
• Further research is needed to improve AI's role in interactive elements and procedural skills training.