Dosimetric effect of respiratory motion prediction error on dynamic multileaf collimator based 4D radiation delivery

Abstract

Purpose/ObjectiveSynchronization of dynamic multileaf collimator response with respiratory motion is critical to ensure the accuracy of 4D radiation delivery. In practice however, a finite time delay (response time RT) between acquisition of tumor position and multileaf collimator (MLC) response necessitates predictive models of respiratory tumor motion to synchronize radiation delivery. Predicting a complex process such as respiratory motion introduces errors that have already been quantified. However, the dosimetric effect of such prediction errors on 4D radiation delivery has not been investigated and quantification of such dosimetric effects forms the subject of this work.Materials/MethodsConformal and IMRT plans for a lung patient were generated for AP-PA geometry at 6 and 18 MV energies. Respiratory motion data was obtained from 60 diaphragm motion recordings of 5 patients. A linear adaptive filter was employed to predict the position of the tumor for 0–0.6 seconds RT. Prediction error was defined as the absolute difference between predicted and actual positions at each diaphragm position. Distributions of prediction error and actual respiratory motion were obtained according to breathing training type (free breathing/audio instructions/visual feedback). Individual dose distributions were convolved with prediction error and respiratory motion distributions. The dosimetric effect of prediction error, along a central plane dose profile was determined as a function of beam energy (6/18 MV), treatment type (conformal/IMRT), beam direction (AP/PA), breathing training modality (free breathing/audio instructions/visual feedback) and RT (0–0.6 seconds). Dose difference was quantified by calculating the maximum and RMS dose error values between each set of convolved and original distributions, expressed as a percentage of maximum actual dose in the original distribution.ResultsIn general, as RT increased, both maximum and RMS dose error increased, indicating smaller delivery errors due to prediction at smaller RT (RMS error dose: 0.2, 0.7, 1.2% at 0.2, 0.4, 0.6 seconds RT for 6 MV conformal and 0.4, 1.2, 2.1% at 0.2, 0.4, 0.6 seconds RT for 6 MV IMRT beam). Direct convolution of the corresponding dose distribution with actual respiratory motion yielded an RMS dose error of 2.2% for the conformal beam and 4% for the IMRT beam. Maximum and RMS dose error were smaller for 18 MV beams (3.8, 0.9% respectively at RT=0.6 seconds, free breathing) than 6 MV beams (4.8, 1.2% of respectively at RT=0.6 seconds, free breathing). Such a decrease is attributed to the larger electron spread in lung for the 18 MV beam energy. Effects of prediction error were more pronounced for IMRT beams as compared to conformal beams (maximum and RMS dose errors of 4.8%, 1.2% for conformal vs. 11.4%, 2.1% for IMRT, 6MV, free breathing, 0.6 seconds RT), due to non-uniform IMRT dose distributions. Breathing training reflected similar trends for dosimetric effects due to prediction error. However, due to larger prediction error with increased respiration amplitude associated with audio instructions, maximum dose errors of up to 8% (conformal) and 16% (IMRT) were observed.ConclusionsPredicting respiratory motion during 4D radiation delivery introduces dosimetric errors that are dependent on several factors, most importantly the response time. Even for relatively small response times of 0.6 seconds into the future, dosimetric errors due to prediction could approach delivery errors when respiratory motion is not accounted for at all. With current prediction methods, response times for 4D radiation delivery should be less than 0.4 seconds. Alternatively, better prediction models would facilitate 4D radiation delivery for longer response times Purpose/ObjectiveSynchronization of dynamic multileaf collimator response with respiratory motion is critical to ensure the accuracy of 4D radiation delivery. In practice however, a finite time delay (response time RT) between acquisition of tumor position and multileaf collimator (MLC) response necessitates predictive models of respiratory tumor motion to synchronize radiation delivery. Predicting a complex process such as respiratory motion introduces errors that have already been quantified. However, the dosimetric effect of such prediction errors on 4D radiation delivery has not been investigated and quantification of such dosimetric effects forms the subject of this work. Synchronization of dynamic multileaf collimator response with respiratory motion is critical to ensure the accuracy of 4D radiation delivery. In practice however, a finite time delay (response time RT) between acquisition of tumor position and multileaf collimator (MLC) response necessitates predictive models of respiratory tumor motion to synchronize radiation delivery. Predicting a complex process such as respiratory motion introduces errors that have already been quantified. However, the dosimetric effect of such prediction errors on 4D radiation delivery has not been investigated and quantification of such dosimetric effects forms the subject of this work. Materials/MethodsConformal and IMRT plans for a lung patient were generated for AP-PA geometry at 6 and 18 MV energies. Respiratory motion data was obtained from 60 diaphragm motion recordings of 5 patients. A linear adaptive filter was employed to predict the position of the tumor for 0–0.6 seconds RT. Prediction error was defined as the absolute difference between predicted and actual positions at each diaphragm position. Distributions of prediction error and actual respiratory motion were obtained according to breathing training type (free breathing/audio instructions/visual feedback). Individual dose distributions were convolved with prediction error and respiratory motion distributions. The dosimetric effect of prediction error, along a central plane dose profile was determined as a function of beam energy (6/18 MV), treatment type (conformal/IMRT), beam direction (AP/PA), breathing training modality (free breathing/audio instructions/visual feedback) and RT (0–0.6 seconds). Dose difference was quantified by calculating the maximum and RMS dose error values between each set of convolved and original distributions, expressed as a percentage of maximum actual dose in the original distribution. Conformal and IMRT plans for a lung patient were generated for AP-PA geometry at 6 and 18 MV energies. Respiratory motion data was obtained from 60 diaphragm motion recordings of 5 patients. A linear adaptive filter was employed to predict the position of the tumor for 0–0.6 seconds RT. Prediction error was defined as the absolute difference between predicted and actual positions at each diaphragm position. Distributions of prediction error and actual respiratory motion were obtained according to breathing training type (free breathing/audio instructions/visual feedback). Individual dose distributions were convolved with prediction error and respiratory motion distributions. The dosimetric effect of prediction error, along a central plane dose profile was determined as a function of beam energy (6/18 MV), treatment type (conformal/IMRT), beam direction (AP/PA), breathing training modality (free breathing/audio instructions/visual feedback) and RT (0–0.6 seconds). Dose difference was quantified by calculating the maximum and RMS dose error values between each set of convolved and original distributions, expressed as a percentage of maximum actual dose in the original distribution. ResultsIn general, as RT increased, both maximum and RMS dose error increased, indicating smaller delivery errors due to prediction at smaller RT (RMS error dose: 0.2, 0.7, 1.2% at 0.2, 0.4, 0.6 seconds RT for 6 MV conformal and 0.4, 1.2, 2.1% at 0.2, 0.4, 0.6 seconds RT for 6 MV IMRT beam). Direct convolution of the corresponding dose distribution with actual respiratory motion yielded an RMS dose error of 2.2% for the conformal beam and 4% for the IMRT beam. Maximum and RMS dose error were smaller for 18 MV beams (3.8, 0.9% respectively at RT=0.6 seconds, free breathing) than 6 MV beams (4.8, 1.2% of respectively at RT=0.6 seconds, free breathing). Such a decrease is attributed to the larger electron spread in lung for the 18 MV beam energy. Effects of prediction error were more pronounced for IMRT beams as compared to conformal beams (maximum and RMS dose errors of 4.8%, 1.2% for conformal vs. 11.4%, 2.1% for IMRT, 6MV, free breathing, 0.6 seconds RT), due to non-uniform IMRT dose distributions. Breathing training reflected similar trends for dosimetric effects due to prediction error. However, due to larger prediction error with increased respiration amplitude associated with audio instructions, maximum dose errors of up to 8% (conformal) and 16% (IMRT) were observed. In general, as RT increased, both maximum and RMS dose error increased, indicating smaller delivery errors due to prediction at smaller RT (RMS error dose: 0.2, 0.7, 1.2% at 0.2, 0.4, 0.6 seconds RT for 6 MV conformal and 0.4, 1.2, 2.1% at 0.2, 0.4, 0.6 seconds RT for 6 MV IMRT beam). Direct convolution of the corresponding dose distribution with actual respiratory motion yielded an RMS dose error of 2.2% for the conformal beam and 4% for the IMRT beam. Maximum and RMS dose error were smaller for 18 MV beams (3.8, 0.9% respectively at RT=0.6 seconds, free breathing) than 6 MV beams (4.8, 1.2% of respectively at RT=0.6 seconds, free breathing). Such a decrease is attributed to the larger electron spread in lung for the 18 MV beam energy. Effects of prediction error were more pronounced for IMRT beams as compared to conformal beams (maximum and RMS dose errors of 4.8%, 1.2% for conformal vs. 11.4%, 2.1% for IMRT, 6MV, free breathing, 0.6 seconds RT), due to non-uniform IMRT dose distributions. Breathing training reflected similar trends for dosimetric effects due to prediction error. However, due to larger prediction error with increased respiration amplitude associated with audio instructions, maximum dose errors of up to 8% (conformal) and 16% (IMRT) were observed. ConclusionsPredicting respiratory motion during 4D radiation delivery introduces dosimetric errors that are dependent on several factors, most importantly the response time. Even for relatively small response times of 0.6 seconds into the future, dosimetric errors due to prediction could approach delivery errors when respiratory motion is not accounted for at all. With current prediction methods, response times for 4D radiation delivery should be less than 0.4 seconds. Alternatively, better prediction models would facilitate 4D radiation delivery for longer response times Predicting respiratory motion during 4D radiation delivery introduces dosimetric errors that are dependent on several factors, most importantly the response time. Even for relatively small response times of 0.6 seconds into the future, dosimetric errors due to prediction could approach delivery errors when respiratory motion is not accounted for at all. With current prediction methods, response times for 4D radiation delivery should be less than 0.4 seconds. Alternatively, better prediction models would facilitate 4D radiation delivery for longer response times