Error detection using a convolutional neural network with dose difference maps in patient-specific quality assurance for volumetric modulated arc therapy

Yuto Kimura, Noriyuki Kadoya, Seiji Tomori, Yohei Oku, Keiichi Jingu

Research output: Contribution to journalArticlepeer-review

13 Citations (Scopus)

Abstract

The aim of this study was to evaluate the use of dose difference maps with a convolutional neural network (CNN) to detect multi-leaf collimator (MLC) positional errors in patient-specific quality assurance for volumetric modulated radiation therapy (VMAT). A cylindrical three-dimensional detector (Delta4, ScandiDos, Uppsala, Sweden) was used to measure 161 beams from 104 clinical prostate VMAT plans. For the simulation used error-free plans plus plans with two types of MLC error were introduced: systematic error and random error. A total of 483 dose distributions in a virtual cylindrical phantom were calculated with a treatment planning system. Dose difference maps were created from two planar dose distributions from the measured and calculated dose distributions, and these were used as the input for the CNN, with 375 datasets assigned for training and 108 datasets assigned for testing. The CNN model had three convolution layers and was trained with five-fold cross-validation. The CNN model classified the error types of the plans as “error-free,” “systematic error,” or “random error,” with an overall accuracy of 0.944. The sensitivity values for the “error-free,” “systematic error,” and “random error” classifications were 0.889, 1.000, and 0.944, respectively, and the specificity values were 0.986, 0.986, and 0.944, respectively. This approach was superior to those based on gamma analysis. Using dose difference maps with a CNN model may provide an effective solution for detecting MLC errors for patient-specific VMAT quality assurance.

Original languageEnglish
Pages (from-to)57-64
Number of pages8
JournalPhysica Medica
Volume73
DOIs
Publication statusPublished - 2020 May

Keywords

  • Deep learning
  • Patient-specific QA
  • Prostate
  • Radiotherapy
  • Volumetric modulated radiation therapy

ASJC Scopus subject areas

  • Biophysics
  • Radiology Nuclear Medicine and imaging
  • Physics and Astronomy(all)

Fingerprint

Dive into the research topics of 'Error detection using a convolutional neural network with dose difference maps in patient-specific quality assurance for volumetric modulated arc therapy'. Together they form a unique fingerprint.

Cite this