Scholarlink

Purpose <p>Accurate Cerebral Blood Flow (CBF) measurements are essential for studying pediatric cerebral hemodynamics. Phase Contrast (PC) imaging is a fast, non-invasive, and non-radiating technique for measuring flows. PC image processing traditionally includes manually segmenting, identifying, and unaliasing vessels of interest, which are challenging in children and involve intra- and inter-observer variation.</p> Methods <p>We acquired 3&#xa0;T PC images from 59 children and 39 adults (mean and standard deviation 3.43 ± 2.60 and 57.28 ± 3.87&#xa0;years). Our algorithm identified voxels that skew the PC image intensity, refined vessels with active contours (Chan-Vese), split adjacent vessels with watershedding, and used a heuristic to identify the correct arteries based on vessel characteristics. We developed an automated algorithm to process PC images, thereby ensuring high precision. Images were processed images manually (two analysts) and algorithmically to compare performance overall and for each component.</p> Results <p>Total CBF measurements were correlated between ground truth and algorithm versus between two analysts (Intraclass Correlation Coefficient = 0.748—0.817 vs 0.810—0.919). A two-way analysis of variance indicated no difference between human and algorithm for total CBF (p = 0.1558). The performance of algorithm versus two human analysts were similar across components: segmentation (Matthew’s Correlation Coefficient = 0.777—0.849 vs 0.830—0.890), unaliasing (Mean Absolute Error = 0.355—0.538 vs 0.410—0.555), and vessel identification in adults (Intraclass Correlation Coefficient = 1.000 vs 1.000). Analysts were similar versus algorithm at vessel identification in children (Intraclass Correlation Coefficient = 1.000 vs 0.983).</p> Conclusion <p>Automated algorithm components performed similarly to gold standard manual analysis and ensured high precision.</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

Development and evaluation of an automated phase contrast magnetic resonance imaging algorithm for pediatric and adult cerebral blood flow measurement

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月27日

查看全文
  • Joseph Liu,
  • Isabel Torres,
  • Sudarshan Ranganathan,
  • Bethany L. Sussman,
  • Eamon K. Doyle,
  • Abhishek Karnwal,
  • Samantha T. Nimmo,
  • Benita Tamrazi,
  • Bradley J. De Souza,
  • Peter A. Chiarelli,
  • Meredith N. Braskie,
  • Hussein N. Yassine,
  • John C. Wood,
  • Bradley S. Peterson,
  • Matthew T. Borzage

摘要

Purpose

Accurate Cerebral Blood Flow (CBF) measurements are essential for studying pediatric cerebral hemodynamics. Phase Contrast (PC) imaging is a fast, non-invasive, and non-radiating technique for measuring flows. PC image processing traditionally includes manually segmenting, identifying, and unaliasing vessels of interest, which are challenging in children and involve intra- and inter-observer variation.

Methods

We acquired 3 T PC images from 59 children and 39 adults (mean and standard deviation 3.43 ± 2.60 and 57.28 ± 3.87 years). Our algorithm identified voxels that skew the PC image intensity, refined vessels with active contours (Chan-Vese), split adjacent vessels with watershedding, and used a heuristic to identify the correct arteries based on vessel characteristics. We developed an automated algorithm to process PC images, thereby ensuring high precision. Images were processed images manually (two analysts) and algorithmically to compare performance overall and for each component.

Results

Total CBF measurements were correlated between ground truth and algorithm versus between two analysts (Intraclass Correlation Coefficient = 0.748—0.817 vs 0.810—0.919). A two-way analysis of variance indicated no difference between human and algorithm for total CBF (p = 0.1558). The performance of algorithm versus two human analysts were similar across components: segmentation (Matthew’s Correlation Coefficient = 0.777—0.849 vs 0.830—0.890), unaliasing (Mean Absolute Error = 0.355—0.538 vs 0.410—0.555), and vessel identification in adults (Intraclass Correlation Coefficient = 1.000 vs 1.000). Analysts were similar versus algorithm at vessel identification in children (Intraclass Correlation Coefficient = 1.000 vs 0.983).

Conclusion

Automated algorithm components performed similarly to gold standard manual analysis and ensured high precision.