biostudies-literatureUnknown15(1)Huang P<h4>Background</h4>Volumetric modulated arc therapy (VMAT) is a popular radiotherapy technique in the clinic. As consisting of hundreds of control points in a VMAT plan it is more complex and time consuming than those conventional treatment modalities, such as intensity modulated radiation therapy. To improve the efficiency and accuracy of its quality assurance procedure, a novel automated anomaly detection method was proposed.<h4>Methods</h4>The anomaly detection model was the Vanilla AutoEncoder (AE). The input was the aperture-based feature maps extracted from the VMAT treatment plan. The output was the reconstruction error in measuring the original and reconstructed feature maps via the low-dimensional latent variables in the bottleneck of the AE model. The AE model was first trained with feature maps extracted from regular plans, and a detection threshold alpha (α) over the distribution of reconstruction errors was then determined. If a larger reconstruction error was obtained for the testing plan, it was considered an anomaly. The data of VMAT plans of 677 patients undergoing lung cancer radiotherapy were collected and tested. The proposed AE was compared with four other classic detection models (principal components analysis, isolation forest, local outlier factor, and hierarchical density-based spatial clustering of applications with noise) using the testing set. To evaluate its reliability, two types of perturbation factors [leaf offset and monitor unit (MU)] were assessed.<h4>Results</h4>Among the five tested models, the AE model achieved the best performance with the area under the receiver operating characteristic curve equal to 0.943. The accuracy and precision of the AE model were 0.769 and 0.407, respectively, which were the highest among the five models. Additionally, in terms of reliability, the AE model was more sensitive in detecting leaf offset and less sensitive in detecting MU variation.<h4>Conclusions</h4>In the automatic physics review of radiotherapy plans, the application of two-dimensional aperture-based feature maps to detect irregular VMAT plans via the AE model is both viable and effective for lung cancer patients.Quantitative imaging in medicine and surgery636-647https://www.ebi.ac.uk/biostudies/studies/S-EPMC11744153biostudies-literatureApplying aperture-based intensity map in automated plan review of volumetric modulated arc therapy for lung cancer patients.PMC11744153Fan YHu ZShang JChang ZDai JYan HHuang PXu YfalseApplying aperture-based intensity map in automated plan review of volumetric modulated arc therapy for lung cancer patients.<h4>Background</h4>Volumetric modulated arc therapy (VMAT) is a popular radiotherapy technique in the clinic. As consisting of hundreds of control points in a VMAT plan it is more complex and time consuming than those conventional treatment modalities, such as intensity modulated radiation therapy. To improve the efficiency and accuracy of its quality assurance procedure, a novel automated anomaly detection method was proposed.<h4>Methods</h4>The anomaly detection model was the Vanilla AutoEncoder (AE). The input was the aperture-based feature maps extracted from the VMAT treatment plan. The output was the reconstruction error in measuring the original and reconstructed feature maps via the low-dimensional latent variables in the bottleneck of the AE model. The AE model was first trained with feature maps extracted from regular plans, and a detection threshold alpha (α) over the distribution of reconstruction errors was then determined. If a larger reconstruction error was obtained for the testing plan, it was considered an anomaly. The data of VMAT plans of 677 patients undergoing lung cancer radiotherapy were collected and tested. The proposed AE was compared with four other classic detection models (principal components analysis, isolation forest, local outlier factor, and hierarchical density-based spatial clustering of applications with noise) using the testing set. To evaluate its reliability, two types of perturbation factors [leaf offset and monitor unit (MU)] were assessed.<h4>Results</h4>Among the five tested models, the AE model achieved the best performance with the area under the receiver operating characteristic curve equal to 0.943. The accuracy and precision of the AE model were 0.769 and 0.407, respectively, which were the highest among the five models. Additionally, in terms of reliability, the AE model was more sensitive in detecting leaf offset and less sensitive in detecting MU variation.<h4>Conclusions</h4>In the automatic physics review of radiotherapy plans, the application of two-dimensional aperture-based feature maps to detect irregular VMAT plans via the AE model is both viable and effective for lung cancer patients.2025-01-01T00:00:00Z2025 Jan2025-04-03T23:58:08.139Z2025-04-03T23:58:08.139ZS-EPMC117441533983899110.21037/qims-24-1398