Scholarlink

Background <p>To evaluate the preliminary feasibility of using generative artificial intelligence (GenAI) to determine optimal entry angles for CT-guided liver needle biopsies in a controlled retrospective setting.</p> Materials and methods <p>This retrospective IRB-approved pilot study analyzed 30 de-identified axial CT images from consecutive liver biopsies performed by 16 interventional radiologists at a single academic center. GenAI operated in 2D on the selected images without specific training, using standardized prompts and overlaid radial grid to generate theoretical trajectory recommendations. Two conditions were evaluated: one with lesions marked and structures to avoid explicitly delineated, and one with lesions marked without annotations. Theoretical trajectory safety assessments were performed by two independent reviewers using consensus methodology. GenAI trajectories were compared retrospectively with clinician-selected paths using intention-to-treat analysis for safety and per-protocol analysis for length.</p> Results <p>In condition one, median iterations were 2 (IQR, 1–3.8), with 27/30 (90%; 95% CI, 74.4–96.9%) GenAI theoretical trajectories rated potentially safe on 2D assessment. Median liver tissue traversed was 31.6&#xa0;mm (IQR, 18.7–67.3&#xa0;mm) versus 48.25&#xa0;mm (IQR, 29.3–61.7&#xa0;mm) for manual paths (<i>p</i> = 0.49). In condition two, median iterations were 1 (IQR, 1–3), with 28/30 (93%; 95% CI, 78.7–98.2%) trajectories rated potentially safe and median liver tissue traversed of 39.35&#xa0;mm (IQR, 24.7–61.6&#xa0;mm) (<i>p</i> = 0.16).</p> Conclusion <p>This pilot study demonstrates limited preliminary feasibility of GenAI for theoretical 2D liver biopsy trajectory planning. These results serve only as proof-of-concept that GenAI can process 2D medical images and generate angle recommendations, not yet as evidence of clinical utility or safety.</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 of generative AI for CT-guided liver biopsy trajectory planning: a pilot study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月26日

查看全文
  • Isabelle Echelman,
  • Taylor Hoffman,
  • Elena N. Petre,
  • Stephen B. Solomon,
  • Francois H. Cornelis

摘要

Background

To evaluate the preliminary feasibility of using generative artificial intelligence (GenAI) to determine optimal entry angles for CT-guided liver needle biopsies in a controlled retrospective setting.

Materials and methods

This retrospective IRB-approved pilot study analyzed 30 de-identified axial CT images from consecutive liver biopsies performed by 16 interventional radiologists at a single academic center. GenAI operated in 2D on the selected images without specific training, using standardized prompts and overlaid radial grid to generate theoretical trajectory recommendations. Two conditions were evaluated: one with lesions marked and structures to avoid explicitly delineated, and one with lesions marked without annotations. Theoretical trajectory safety assessments were performed by two independent reviewers using consensus methodology. GenAI trajectories were compared retrospectively with clinician-selected paths using intention-to-treat analysis for safety and per-protocol analysis for length.

Results

In condition one, median iterations were 2 (IQR, 1–3.8), with 27/30 (90%; 95% CI, 74.4–96.9%) GenAI theoretical trajectories rated potentially safe on 2D assessment. Median liver tissue traversed was 31.6 mm (IQR, 18.7–67.3 mm) versus 48.25 mm (IQR, 29.3–61.7 mm) for manual paths (p = 0.49). In condition two, median iterations were 1 (IQR, 1–3), with 28/30 (93%; 95% CI, 78.7–98.2%) trajectories rated potentially safe and median liver tissue traversed of 39.35 mm (IQR, 24.7–61.6 mm) (p = 0.16).

Conclusion

This pilot study demonstrates limited preliminary feasibility of GenAI for theoretical 2D liver biopsy trajectory planning. These results serve only as proof-of-concept that GenAI can process 2D medical images and generate angle recommendations, not yet as evidence of clinical utility or safety.