MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A Novel Prediction of MLC Leaf Positioning Accuracy Based On Artificial Neural Network

B Min1*, H Nam1 , W Cheon2 , Y Han2 , H Lee1 , Y Kim1 , I Jeong1 , S Bang1 , J Park1 , (1) Department of Radiation Oncology, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul (2) Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul

Presentations

(Sunday, 7/29/2018) 3:00 PM - 6:00 PM

Room: Exhibit Hall

Purpose: The volumetric modulated arc therapy (VMAT) and intensity modulated radiation therapy (IMRT) have been introduced for photon therapy using multi-leaf collimators (MLC) delivering maximum dose to tumor tissue while minimizing the absorbed dose in healthy tissue. A multileaf collimator (MLC) leaf position uncertainties directly affect the dose delivered to tumor targets and sensitive structures in IMRT and VMAT. Therefore, it is essential to conduct precise MLC quality assurance (QA) for position and speed of each leaf. In this study, we propose MLC pre-QA method using artificial neural network affordable to time sequence data for prediction of mechanical error of each leaf in MLC.

Methods: RTpaln dicom file which has information about the treatment plan was converted to Dynalog expected position by in-house program for the estimation of MLC mechanical error before the radiation treatment. Second, we construct artificial neural network for predicting actual position of MLC reflecting mechanical error of linear accelerator (Clinac: Varian, Palo Alto, California, United States). The neural network consist of 5 long short term memory (LSTM) cell (4 gates, 128 neurons) and fully connected layers (1024 neuron) developed using open-source software library for machine intelligence (Tensorflow). Finally, the error of predicted actual positions of MLC is evaluated through maximum likely hood function.

Results: We build workflow for predicting mechanical errors of MLC before radiation treatment. The evaluation of LSTM network is performed by test set of Dynalog files. The accuracy of training and test set were 98% and 93%, respectively.

Conclusion: This work indicates MLC Pre-QA method based on artificial neural network showed the prediction of mechanical error of MLC and could provide the probability of error that could occur during the treatment. This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIP) (No. NRF-2017R1C1B2011257)

Keywords

Quality Assurance, Intensity Modulation, Treatment Verification

Taxonomy

TH- External beam- photons: Quality Assurance - VMAT

Contact Email