Scholarlink

Background <p>Quantification of myocardial blood flow (MBF) with <InlineEquation ID="IEq3"><EquationSource Format="TEX">\(^{82}\)</EquationSource></InlineEquation>Rb PET/CT requires accurate delineation of the left ventricle (LV). Manual or semi-automated contouring remains time-consuming and error-prone, particularly in hypoperfused myocardium. We developed and validated a fully automatic LV segmentation pipeline using nnU-Net applied to <InlineEquation ID="IEq4"><EquationSource Format="TEX">\(^{82}\)</EquationSource></InlineEquation>Rb PET/CT. A manual, multimodal segmentation protocol integrating dynamic PET and CT was established in a single-center cohort of 40 non-gated PET/CT series (20 patients, rest &amp; stress), including challenging cases with extensive necrosis (median 21%). The resulting ground truth masks were used for five-fold cross-validation, and semi-supervised learning incorporated 805 additional unlabeled dynamic PET series (504 patients). Model performance was compared with an optimized semi-automatic thresholding baseline (35% <InlineEquation ID="IEq5"><EquationSource Format="TEX">\(\text {SUV}_{\text {max}}\)</EquationSource></InlineEquation>).</p> Results <p>The nnU-Net significantly outperformed the baseline, achieving a mean Dice of 87.8[85.6, 89.2]% vs 75.1[72.9, 76.9]%, recall 89.1[86.1, 91.4]% vs 82.6[79.1, 85.4]%, and precision 88.1[84.2, 90.4]% vs 70.2[67.2, 73.0]%. The improvement was most pronounced in hypoperfused regions, where recall increased by 20–30% compared to thresholding. Semi-supervised learning modestly enhanced model robustness across both rest and stress acquisitions.</p> Conclusions <p>A deep-learning-based approach enables fully automatic LV segmentation in <InlineEquation ID="IEq6"><EquationSource Format="TEX">\(^{82}\)</EquationSource></InlineEquation>Rb PET/CT with near-expert accuracy. This framework eliminates manual interaction, supports large-scale MBF quantification, and paves the way for reproducible, high-throughput cardiac PET analysis in clinical and research workflows.</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

Fully automatic left ventricle segmentation in \(^{82}\)Rb PET/CT Using a semi-supervised nnU-net

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Mohammadreza Amirian,
  • Arthur Chevalley,
  • María Martín Asiain,
  • Ran Klein,
  • Robert DeKemp,
  • Eric Moulton,
  • John O. Prior,
  • Christel H. Kamani,
  • Mario Jreige,
  • Adrien Depeursinge

摘要

Background

Quantification of myocardial blood flow (MBF) with \(^{82}\)Rb PET/CT requires accurate delineation of the left ventricle (LV). Manual or semi-automated contouring remains time-consuming and error-prone, particularly in hypoperfused myocardium. We developed and validated a fully automatic LV segmentation pipeline using nnU-Net applied to \(^{82}\)Rb PET/CT. A manual, multimodal segmentation protocol integrating dynamic PET and CT was established in a single-center cohort of 40 non-gated PET/CT series (20 patients, rest & stress), including challenging cases with extensive necrosis (median 21%). The resulting ground truth masks were used for five-fold cross-validation, and semi-supervised learning incorporated 805 additional unlabeled dynamic PET series (504 patients). Model performance was compared with an optimized semi-automatic thresholding baseline (35% \(\text {SUV}_{\text {max}}\)).

Results

The nnU-Net significantly outperformed the baseline, achieving a mean Dice of 87.8[85.6, 89.2]% vs 75.1[72.9, 76.9]%, recall 89.1[86.1, 91.4]% vs 82.6[79.1, 85.4]%, and precision 88.1[84.2, 90.4]% vs 70.2[67.2, 73.0]%. The improvement was most pronounced in hypoperfused regions, where recall increased by 20–30% compared to thresholding. Semi-supervised learning modestly enhanced model robustness across both rest and stress acquisitions.

Conclusions

A deep-learning-based approach enables fully automatic LV segmentation in \(^{82}\)Rb PET/CT with near-expert accuracy. This framework eliminates manual interaction, supports large-scale MBF quantification, and paves the way for reproducible, high-throughput cardiac PET analysis in clinical and research workflows.