Scholarlink

Background <p>Accurate target delineation is essential for achieving optimal plan quality in craniospinal irradiation (CSI), particularly in pediatric proton therapy. However, manual target delineation is highly time-consuming due to the large anatomical extent of craniospinal axis. This study aimed to develop and evaluate a deep learning-based automated segmentation model to streamline the target delineation for pediatric proton CSI.</p> Methods <p>Fifty pediatric patient datasets treated with proton CSI at the National Cancer Center Korea were utilized. The nnU-Net v2 framework was employed for model training using 40 datasets, with the remaining 10 used for independent evaluation. Segmentation performance was assessed using the Dice Similarity Coefficient (DSC), Intersection over Union (IoU), 95th percentile Hausdorff Distance (HD95), and Average Symmetric Surface Distance (ASSD). Workflow efficiency was evaluated by comparing manual and automated contouring times, and clinical acceptability was assessed using a Likert scale.</p> Results <p>The model achieved high geometric accuracy, with mean DSC, IoU, HD95, and ASSD values of 0.9587, 0.9222, 2.4948&#xa0;mm, and 0.7900&#xa0;mm, respectively. The target delineation time was reduced from approximately 8&#xa0;h to 30&#xa0;min. Clinical acceptability was favorable, with the majority of Likert scale ratings falling within the positive-to-neutral range.</p> Conclusions <p>This study demonstrates that the developed deep learning-based automated segmentation model can achieve accurate and robust target delineation for proton CSI. By integrating this framework into the planning workflow, CSI can be performed with greater efficiency, reproducibility, and standardization. Our findings highlight the potential of automation to advance precision and productivity in pediatric proton CSI practice.</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

Deep learning-based automated segmentation of target volumes for pediatric craniospinal irradiation in proton therapy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月31日

查看全文
  • Sung Hun Kim,
  • Geon Oh,
  • Joo-Young Kim,
  • Se Byeong Lee,
  • Dongho Shin,
  • Young Kyung Lim,
  • Haksoo Kim,
  • Jong Hwi Jeong,
  • Chankyu Kim

摘要

Background

Accurate target delineation is essential for achieving optimal plan quality in craniospinal irradiation (CSI), particularly in pediatric proton therapy. However, manual target delineation is highly time-consuming due to the large anatomical extent of craniospinal axis. This study aimed to develop and evaluate a deep learning-based automated segmentation model to streamline the target delineation for pediatric proton CSI.

Methods

Fifty pediatric patient datasets treated with proton CSI at the National Cancer Center Korea were utilized. The nnU-Net v2 framework was employed for model training using 40 datasets, with the remaining 10 used for independent evaluation. Segmentation performance was assessed using the Dice Similarity Coefficient (DSC), Intersection over Union (IoU), 95th percentile Hausdorff Distance (HD95), and Average Symmetric Surface Distance (ASSD). Workflow efficiency was evaluated by comparing manual and automated contouring times, and clinical acceptability was assessed using a Likert scale.

Results

The model achieved high geometric accuracy, with mean DSC, IoU, HD95, and ASSD values of 0.9587, 0.9222, 2.4948 mm, and 0.7900 mm, respectively. The target delineation time was reduced from approximately 8 h to 30 min. Clinical acceptability was favorable, with the majority of Likert scale ratings falling within the positive-to-neutral range.

Conclusions

This study demonstrates that the developed deep learning-based automated segmentation model can achieve accurate and robust target delineation for proton CSI. By integrating this framework into the planning workflow, CSI can be performed with greater efficiency, reproducibility, and standardization. Our findings highlight the potential of automation to advance precision and productivity in pediatric proton CSI practice.