Scholarlink

Background <p>Precise modeling of boron distribution is essential for boron neutron capture therapy (BNCT) dose calculation. While full-three-dimensional (f3D) voxel modeling of ¹⁸F-BPA PET data offers high spatial resolution, its clinical value is limited by intrinsic image resolution—especially PET—and by registration errors in multi-modality image fusion. These factors constrain the reliability of voxel-level dose interpretation and increase computational burden. We propose a hypo-segmentation (hS) method that groups target tissue-to-blood ratio (TBR) values into 0.5-unit intervals to approximate boron distribution. This study retrospectively evaluated four malignant brain tumor cases, comparing dosimetric metrics derived from hS and f3D approaches. Key dose-volume parameters (D<sub>mean</sub>, D<sub>80</sub>, V<sub>14 Gy−Eq</sub>), spatial agreement (gamma pass rate), and dose–volume curve similarity (dynamic time warping, DTW) were assessed. Dose consistency in normal brain tissue was also evaluated using D<sub>0.1 cc</sub>, D<sub>mean</sub>, and V<sub>13 Gy−Eq</sub>.</p> Results <p>The hS method showed strong agreement with f3D results: within 2% for tumor D<sub>mean</sub>, 3% for D<sub>80</sub>, and 5% for V<sub>14 Gy−Eq</sub>. Gamma pass rates exceeded 90% in three of four cases. In brain tissue, dose deviations between hS and f3D remained within 2%, demonstrating reliability even in low-uptake, spatially uniform regions. A single small, heterogeneous tumor showed larger discrepancy (D<sub>95</sub> deviation = 14.3%, gamma pass rate = 82%), which was mitigated by refining segmentation (h = 0.25).</p> Conclusions <p>The hS method offers a clinically practical and computationally efficient alternative to full voxel-wise modeling. It preserves dosimetric fidelity in both tumor and normal brain regions, while aligning with the intrinsic resolution of PET and multimodal imaging. This approach supports accurate yet streamlined BNCT dose planning, particularly in clinical settings where ultra-high spatial granularity provides limited added value.</p> Trial registration number (TRN) <p>ChiCTR2200066473, registered on December 16, 2022.</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

Hypo-segmentation of 18F-BPA PET-based boron distribution: a pragmatic approach for efficient BNCT treatment planning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月8日

查看全文
  • Yi-Chiao Teng,
  • Jiang Chen,
  • Wan-Bing Zhong,
  • Qin Lin,
  • Yuan-Hao Liu

摘要

Background

Precise modeling of boron distribution is essential for boron neutron capture therapy (BNCT) dose calculation. While full-three-dimensional (f3D) voxel modeling of ¹⁸F-BPA PET data offers high spatial resolution, its clinical value is limited by intrinsic image resolution—especially PET—and by registration errors in multi-modality image fusion. These factors constrain the reliability of voxel-level dose interpretation and increase computational burden. We propose a hypo-segmentation (hS) method that groups target tissue-to-blood ratio (TBR) values into 0.5-unit intervals to approximate boron distribution. This study retrospectively evaluated four malignant brain tumor cases, comparing dosimetric metrics derived from hS and f3D approaches. Key dose-volume parameters (Dmean, D80, V14 Gy−Eq), spatial agreement (gamma pass rate), and dose–volume curve similarity (dynamic time warping, DTW) were assessed. Dose consistency in normal brain tissue was also evaluated using D0.1 cc, Dmean, and V13 Gy−Eq.

Results

The hS method showed strong agreement with f3D results: within 2% for tumor Dmean, 3% for D80, and 5% for V14 Gy−Eq. Gamma pass rates exceeded 90% in three of four cases. In brain tissue, dose deviations between hS and f3D remained within 2%, demonstrating reliability even in low-uptake, spatially uniform regions. A single small, heterogeneous tumor showed larger discrepancy (D95 deviation = 14.3%, gamma pass rate = 82%), which was mitigated by refining segmentation (h = 0.25).

Conclusions

The hS method offers a clinically practical and computationally efficient alternative to full voxel-wise modeling. It preserves dosimetric fidelity in both tumor and normal brain regions, while aligning with the intrinsic resolution of PET and multimodal imaging. This approach supports accurate yet streamlined BNCT dose planning, particularly in clinical settings where ultra-high spatial granularity provides limited added value.

Trial registration number (TRN)

ChiCTR2200066473, registered on December 16, 2022.