Scholarlink

Objective <p>To evaluate the feasibility of using generative artificial intelligence (GenAI) in determining entry angles for lung needle biopsies.</p> Materials and methods <p>This retrospective IRB-approved study analyzed de-identified computed tomography (CT) images from 30 lung biopsy procedures using GPT-4 (OpenAI, San Francisco, CA) to identify biopsy paths that minimize tissue disruption and avoid critical structures. A radial grid was overlaid on axial CT images to facilitate identification of entry angles (EA). The proposed GenAI EA were compared with the biopsy trajectories selected by clinicians during manual procedures.</p> Results <p>The median number of iterations required to identify the GenAI EA was 2 (interquartile range: 1–3). Independent reviewers deemed 28 of 30 (93.3%) GenAI-generated paths safe for biopsy. The median distance traveled through lung tissue by the GenAI paths was 5 mm (interquartile range: 0–25 mm), significantly shorter compared to the manually selected paths (23 mm; 5–57 mm; <i>p</i> = 0.0063).</p> Conclusion <p>Generative AI can effectively determine feasible biopsy trajectories, potentially leading to increased standardization in lung biopsy procedure planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Feasibility study on using multimodal large language model for CT-guided lung biopsy trajectory planning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月5日

查看全文
  • Taylor Hoffman,
  • Megan Worthington,
  • Debkumar Sarkar,
  • Stephen B. Solomon,
  • Francois H. Cornelis

摘要

Objective

To evaluate the feasibility of using generative artificial intelligence (GenAI) in determining entry angles for lung needle biopsies.

Materials and methods

This retrospective IRB-approved study analyzed de-identified computed tomography (CT) images from 30 lung biopsy procedures using GPT-4 (OpenAI, San Francisco, CA) to identify biopsy paths that minimize tissue disruption and avoid critical structures. A radial grid was overlaid on axial CT images to facilitate identification of entry angles (EA). The proposed GenAI EA were compared with the biopsy trajectories selected by clinicians during manual procedures.

Results

The median number of iterations required to identify the GenAI EA was 2 (interquartile range: 1–3). Independent reviewers deemed 28 of 30 (93.3%) GenAI-generated paths safe for biopsy. The median distance traveled through lung tissue by the GenAI paths was 5 mm (interquartile range: 0–25 mm), significantly shorter compared to the manually selected paths (23 mm; 5–57 mm; p = 0.0063).

Conclusion

Generative AI can effectively determine feasible biopsy trajectories, potentially leading to increased standardization in lung biopsy procedure planning.