Scholarlink

<p>To predict and classify the gamma passing rate (GPR) for patient-specific quality assurance (PSQA) in VMAT using a deep learning approach based on accelerator dynamic log files. Dynamic log files of 438 VMAT plans were analyzed. GPR were measured by using Delta4. Using dynamic log files as inputs and GPR under five different criteria as outputs, the Resnet-based model was trained by a five-fold cross-validation method. The average absolute error (MAE), mean square error (MSE), root mean square error (RMSE), Spearman’s rank correlation coefficient (Sr), coefficient of determination (R<sup>2</sup>), and Classification performance were calculated. In the test dataset, the predictive model demonstrated good agreement with measured GPR values across different gamma criteria, with MAE ranging from 0.59 to 2.05%, RMSE ranging from 1.02 to 3.94%, and R<sup>2</sup> ranging from 0.76 to 0.88. Significant correlations were observed between predicted and measured GPR values under all gamma criteria (Sr = 0.83–0.94, <i>p</i> &lt; 0.001).For the classification models, the accuracy ranged from 0.87 to 0.96, while the AUC values ranged from 0.89 to 0.97 across different gamma criteria, indicating favorable classification performance. We developed a deep learning model that demonstrated promising performance in predicting and classifying GPR for VMAT PSQA. The model is promising as it streamlines the PSQA process and may reduces the associated workload, demonstrating substantial clinical potential.</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 for patient-specific quality assurance of volumetric modulated arc therapy: predicting gamma passing rate based on accelerator dynamic log files

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Haihua Lei,
  • Te Xu,
  • Qianhong Chen,
  • Zhenhang Xu,
  • Jiaxiang Gao,
  • Jiangshan Li,
  • Yuxin Li,
  • Yihan Liu,
  • Xianwen Li,
  • Xiaodong Wu,
  • Yuangui Chen

摘要

To predict and classify the gamma passing rate (GPR) for patient-specific quality assurance (PSQA) in VMAT using a deep learning approach based on accelerator dynamic log files. Dynamic log files of 438 VMAT plans were analyzed. GPR were measured by using Delta4. Using dynamic log files as inputs and GPR under five different criteria as outputs, the Resnet-based model was trained by a five-fold cross-validation method. The average absolute error (MAE), mean square error (MSE), root mean square error (RMSE), Spearman’s rank correlation coefficient (Sr), coefficient of determination (R2), and Classification performance were calculated. In the test dataset, the predictive model demonstrated good agreement with measured GPR values across different gamma criteria, with MAE ranging from 0.59 to 2.05%, RMSE ranging from 1.02 to 3.94%, and R2 ranging from 0.76 to 0.88. Significant correlations were observed between predicted and measured GPR values under all gamma criteria (Sr = 0.83–0.94, p < 0.001).For the classification models, the accuracy ranged from 0.87 to 0.96, while the AUC values ranged from 0.89 to 0.97 across different gamma criteria, indicating favorable classification performance. We developed a deep learning model that demonstrated promising performance in predicting and classifying GPR for VMAT PSQA. The model is promising as it streamlines the PSQA process and may reduces the associated workload, demonstrating substantial clinical potential.