An automatic patient-specific quality assurance with a novel DVH scoring algorithm for volumetric modulated arc therapy of cervical cancer

• Published in: Applied radiation and isotopes Vol. 225; p. 112030
• Main Authors: Zhang, Ji, Ning, Boda, Jin, Xiaotong, Shen, Yanting, Mu, Yiran, Yi, Jinling, Han, Ce, Zhou, Yongqiang, Bai, Yanling, Jin, Xiance
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.11.2025
• Subjects: Algorithms; Classification; Deep learning; Dose-volume histogram; Female; Humans; Patient-specific quality assurance; Quality Assurance, Health Care; Radiotherapy Dosage; Radiotherapy Planning, Computer-Assisted; Radiotherapy, Intensity-Modulated - methods; Radiotherapy, Intensity-Modulated - standards; Uterine Cervical Neoplasms - radiotherapy; Volumetric modulated arc therapy; Deep learning; Dose-volume histogram; Volumetric modulated arc therapy; Patient-specific quality assurance; Classification
• ISSN: 0969-8043; 1872-9800; 1872-9800
• DOI: 10.1016/j.apradiso.2025.112030

To develop a novel DVH scoring algorithm to predict and classify patient-specific quality assurance (PSQA) results by different DVH metrics automatically and efficiently. A total of 200 cervical cancer patients who treated by Infinity (109 cases) and Synergy (91 cases) linear accelerators underwent volumetric modulated arc therapy (VMAT) from 2019 to 2022 were enrolled and used as technical testing (TT) and technical validation (TV) datasets, respectively, which was then randomly divided into training, validation and testing set at a ratio of 7:1:2. PSQA dose distributions were predicted using U-shape-like network with skip-connection modules (called T-Net) with the input of CT and plan dose distributions. A novel weight-based DVH scoring (WDS) algorithm was developed and trained to classify “pass” or “fail” (PoF) of PSQA results based on the dose errors (DEs) and volumetric errors (VEs) calculated between predicted and planned DVHs. T-Net achieved a best performance in predicting PSQA dose distributions in comparison with other deep learning models. The WDS method achieved a sensitivity, specificity and accuracy of 100.00 %, 50.00 %, 0.955, and 100.00 %,33.33 %, 0.890 in TT and TV, respectively, which was better than models of random forest (RF) and support vector machines (SVM) with an accuracy of 0.909, 0.833 and 0.864, 0.722 in TT and TV, respectively. The threshold DVH score for 22 and 18 validation patients were 49.62 and 57.62 in the TT and TV with a precision, recall rate and F1 score of 0.952, 1, 0.976 and 0.882, 1, 0.938, respectively. The suggested novel WDS algorithm can improve the accuracy and efficiency of classifying the PoF of PSQA objectively and automatically. •A novel patient-specific quality assurance method based DVH was develop and assessed.•The weight-based DVH scoring algorithm is able to classify the result of plans.•The proposed method can be applied in plans delivered on different LINACs.