Evaluating large language model-generated brain MRI protocols: performance of GPT4o, o3-mini, DeepSeek-R1 and Qwen2.5-72B - Takeaways - MDSpire

Evaluating large language model-generated brain MRI protocols: performance of GPT4o, o3-mini, DeepSeek-R1 and Qwen2.5-72B

  • By

  • Su Hwan Kim

  • Severin Schramm

  • Lena Schmitzer

  • Kerem Serguen

  • Sebastian Ziegelmayer

  • Felix Busch

  • Alexander Komenda

  • Marcus R. Makowski

  • Lisa C. Adams

  • Keno K. Bressem

  • Claus Zimmer

  • Jan Kirschke

  • Benedikt Wiestler

  • Dennis Hedderich

  • Tom Finck

  • Jannis Bodden

  • September 3, 2025

  • 0 min

Share

  • 1

    The study evaluates the performance of four large language models in generating brain MRI protocols based on realistic clinical cases.

  • 2

    Protocoling for brain MRI is crucial, balancing comprehensive imaging with efficiency to minimize repeat examinations and healthcare costs.

  • 3

    Two experienced neuroradiologists defined reference protocols for 150 fictitious cases, resolving discrepancies through consensus.

  • 4

    The models tested include GPT-4o, o3-mini, DeepSeek-R1, and Qwen2.5-72B, with evaluations conducted under enhanced and base conditions.

  • 5

    The study aims to assess AI's potential in improving MRI protocoling, addressing the increasing demands on radiologists' time.

Original Source(s)

Evaluating large language model-generated brain MRI protocols: performance of GPT4o, o3-mini, DeepSeek-R1 and Qwen2.5-72B

  • by Su Hwan Kim, Severin Schramm, Lena Schmitzer, Kerem Serguen, Sebastian Ziegelmayer, Felix Busch, Alexander Komenda, Marcus R. Makowski, Lisa C. Adams, Keno K. Bressem, Claus Zimmer, Jan Kirschke, Benedikt Wiestler, Dennis Hedderich, Tom Finck, Jannis Bodden

  • September 3, 2025

Related Content

A spatial correlation-guided deep fusion framework for multimodal lung cancer classification using CT imaging

Emerging applications of artificial intelligence for risk stratification in head and neck cancer: a scoping review

MRI-based habitat radiomics for preoperative prediction of axillary pathological complete response in breast cancer after neoadjuvant therapy: a multicenter study