JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

The clinical application potential assessment of the Deepseek-R1 large language model in lung cancer

  • 0The Graduate School of Qinghai University, Xining, Qinghai, China.
Frontiers in Oncology +

|

September 18, 2025

|

September 18, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

The large language model Deepseek-R1 demonstrates superior diagnostic accuracy and treatment recommendations compared to junior oncologists in lung cancer care. While ethical considerations require further refinement, Deepseek-R1 shows significant potential to aid physicians.

Area Of Science

  • Oncology
  • Artificial Intelligence
  • Medical Informatics

Background

  • Evaluating the clinical utility of advanced AI, specifically the large language model Deepseek-R1.
  • Focusing on Deepseek-R1's role in assisting junior oncologists with lung cancer diagnosis and treatment.
  • Assessing diagnostic accuracy, treatment consistency, and clinical decision-making reliability.

Purpose Of The Study

  • To compare the performance of Deepseek-R1 against junior oncologists in lung cancer management.
  • To identify the strengths and limitations of AI in clinical oncology decision support.
  • To explore the potential of AI in enhancing the capabilities of early-career physicians.

Main Methods

  • Retrospective analysis of 320 newly diagnosed lung cancer patients.
  • Development of 26 clinical questions based on international guidelines covering knowledge, decision-making, and ethics.
  • Double-blind evaluation of Deepseek-R1 and junior oncologist responses by senior oncologists.

Main Results

  • Deepseek-R1 achieved higher accuracy rates (92.3% basic knowledge, 87.5% complex decisions, 85.1% ethics) than junior oncologists (80.4%, 72.8%, 70.2%).
  • Overall diagnostic accuracy for Deepseek-R1 was 94.6% versus 78.9% for junior oncologists.
  • Deepseek-R1 showed high consistency in treatment updates; junior oncologists had more logical errors, while the model had more ethical risks.

Conclusions

  • Deepseek-R1 significantly surpasses junior oncologists in diagnostic accuracy and treatment decision-making for lung cancer.
  • The model shows promise for supporting junior physicians, facilitating discussions, and optimizing treatment plans.
  • Further development is needed to address ethical reasoning limitations in AI-driven clinical support.

Keywords: