- Email: [email protected]
PO-0848: Towards automated treatment planning in radiotherapy
S80
ESTRO 33, 2014
Aim of the study was that of investigating the parotid Davg increase during IMRT treatments using local pragmatic settings. Materials and Methods: Starting from Jun 2013,we prospectively enrolled patients with SCCHN listed for IMRT treatment. There were no exclusion criteria, to stress the pragmatic conditions of the study (i.e. patients with dental implants, not satisfactory CT scans, etc). For each patient, the treatment plan was calculated according to the routine procedure using the CT simulator (Phase 0 CT scan). V and Davg of each parotid were collected and stored as reference V and Davg. In addition to the therapeutic routine, CT scans of the patient were acquired after half treatment (Phase 1 CT scan) and at the end of the treatment (Phase 2 CT scan) and parotids were contoured by the same radiotherapist. The Treatment Plan calculated in Phase 0 was superimposed to the Phase 1 and Phase 2 CT scans, in order to obtain Phase 1 and Phase 2 values of parotid V and Davg. The reduction in V was calculated as percentage reduction. The variation in Davg (ΔD) was evaluated considering the difference between Phase 0 Davg and the Real Dose (DR), calculated as the arithmetic mean of Phase0, 1 and 2 Davg. Correlation between V reduction and Davg variation was investigated. Results: 6 patients have been enrolled until now. IMRT treatment prescriptions ranged from 54 to 70Gy. V reduction was seen in all the parotids included, varying from 16,3 to 56,0 % (mean 36,5). ΔD was positive for 5 parotids and negative for 7 parotids; for the subgroup of parotids for which Davg < 20 Gy, average ΔD was negative, while for non-preserved parotids (Davg > 50 Gy) average ΔD was positive. For parotids with Davg closer to the 26 Gy threshold, the behaviour of ΔD was ambiguous, with both massive increments (+ 464 cGy starting from 2759 cGy) and decrements (- 427cGy starting from 4571 cGy). There were no correlation between Davg and ΔD (R2= 0,151); lack of correlation was also showed between V and ΔD (R2= 0,148). Patients’ enrolling is still undergoing. Conclusions: Preservation of parotids for SCCHN patients treated with IMRT has to be pursued keeping into account the unpredictable and drastic variation in Davg attributed to the gland. V variation and initial Davgare not good predictors of ΔD.
POSTER: PHYSICS OPTIMISATION
TRACK:
TREATMENT
PLAN
PO-0848 Towards automated treatment planning in radiotherapy S. Breedveld1 1 Erasmus Medical Center Rotterdam, Radiation Oncology, Rotterdam, The Netherlands
compare different coplanar and non-coplanar beam set-ups for conventional linac and CyberKnife. The lack of posterior beams in the CyberKnife system did not significantly influence the achieved plan quality. In another study, adaptive strategies in liver SBRT were investigated. It was demonstrated that automated daily re-optimization of fluence profiles did almost as well as re-optimization of both beam profiles and beam angles. iCycle is now in daily routine clinical use for an expanding number of treatment sites. Conclusions: With iCycle it is possible to automatically generate treatment plans. Sofar, all studies and clinical experience point at a quality that is superior to plans made by dosimetrists and physicists. Because of the deterministic nature of the algorithm, iCycle can be used to objectively compare treatment strategies. PO-0849 A method to estimate sweeping window arc therapy (SWAT) treatment plan robustness J. Fleckenstein1, J. Hesser1, F. Wenz1, F. Lohr1 1 University Medical Center Mannheim, Department for Radiation Oncology and Radiotherapy, Mannheim, Germany Purpose/Objective: In VMAT dose deliveries with the SWAT technique, in contrast to conventional step-and-shoot or dynamic IMRT with discrete gantry angles, the effect of intrafractional motion on the dose distribution needs further evaluation. With a fast MLC and flattening filter free (FFF) dose delivery, treatment times close to 60 s per fraction are clinical reality. For these treatment times the human breathing period gets close to the period of the collimator sweeping period. Therefore methods to a) quantify the effect of intrafractional target motion and b) provide a sensitive parameter during treatment planning to avoid potentially non-robust treatment plans are presented. Materials and Methods: An Elekta Synergy LINAC with MLCi2 MLC and an Elekta VersaHD LINAC with Agility MLC were used for deliveries of 50 clinical treatment sequences. All treatment deliveries were performed with one 360° VMAT arc and were generated with the Elekta Monaco (V. 3.2-3.3) treatment planning system. Measurements were performed with an ionization chamber array (MatriXX Evolution, IBA) inside a Multicube Lite phantom (IBA) with a sampling time of 100 ms. To detect interplay effects along all three coordinate axes, measurements were performed twice with the detector aligned in vertical or horizontal direction. All slices were moved against each other in MATLAB according to a pre-defined motion vector, mimicking a cosine-to-the-square motion. The minimum relative equivalent uniform dose (EUD) for all breathing starting phases Φ was determined for every amplitude A and period T.
Purpose/Objective: Treatment plan optimization is a labour-intensive time-consuming task, with quality highly dependent on the skills and experience of the planner. Combined with the multi-objective origin of the problem and large degrees of freedom in treatment device setup, the resulting treatment plan may be suboptimal. To overcome these limitations, we developed an algorithm, iCycle, to fully automate treatment planning.
Furthermore LINAC logfiles were recorded during plan delivery. The MLC, jaw, gantry angle and monitor unit settings were continuously saved and used to calculate the cross correlation coefficient between the target motion and the collimator motion component for each direction (CC, LR, AP) separately. The cross correlation coefficient ρ of the dose weighed collimator positions xc and the breathing positions xt was expressed as a product of the standard scores as:
Materials and Methods: To automate the decision making process in treatment planning, iCycle uses lexicographic optimization where hard constraints and prioritized treatment objectives are summarized in a socalled wish-list. Constraints are met unconditionally, while the objectives are optimized in priority. The focus lies on attaining the treatment goals (in priority), where a Pareto optimal plan is obtained after maximally optimizing all objectives. iCycle has an advanced option for coplanar and non-coplanar beam angle selection, integrated with IMRT beam profile optimization. A wish-list is tuned in a collaboration between clinicians and dosimetrists, to be used for a group of similar patients, e.g. all prostate patients with radical treatment. There is no need for adaptation for individual patients. Due to the automation, iCycle is highly suited for objectively comparing treatment strategies, based on large numbers of automatically generated treatment plans.
Results: Figure 1 shows the minimum relative EUD (upper figure) in comparison to the cross correlation coefficient (lower figure) as functions of the period. In the presented case motion in left-right direction was applied. Plan parameters were: Treatment time 102 s, Agility MLC, 20°increment, 179 control points, 2.0 Gy fraction dose, 655.7 MU. The minimum relative EUD observed in all cases was 94.6%. For all cases a cross correlation coefficient above 0.5 corresponded to a minimum in EUD.
Results: iCycle was used for several studies. In a prospective clinical head-and-neck study, both manually generated plans and plans generated by iCycle were blindly presented to the treating clinician. In 32/33 cases, the treating clinician selected the plan generated by iCycle. A treatment planning study for prostate cancer patients with one or two metal hip prosthesis resulted in recommandations for complex cases. For prostate SBRT, iCycle generated 1500 plans automatically to
ESTRO 33, 2014
Aim of the study was that of investigating the parotid Davg increase during IMRT treatments using local pragmatic settings. Materials and Methods: Starting from Jun 2013,we prospectively enrolled patients with SCCHN listed for IMRT treatment. There were no exclusion criteria, to stress the pragmatic conditions of the study (i.e. patients with dental implants, not satisfactory CT scans, etc). For each patient, the treatment plan was calculated according to the routine procedure using the CT simulator (Phase 0 CT scan). V and Davg of each parotid were collected and stored as reference V and Davg. In addition to the therapeutic routine, CT scans of the patient were acquired after half treatment (Phase 1 CT scan) and at the end of the treatment (Phase 2 CT scan) and parotids were contoured by the same radiotherapist. The Treatment Plan calculated in Phase 0 was superimposed to the Phase 1 and Phase 2 CT scans, in order to obtain Phase 1 and Phase 2 values of parotid V and Davg. The reduction in V was calculated as percentage reduction. The variation in Davg (ΔD) was evaluated considering the difference between Phase 0 Davg and the Real Dose (DR), calculated as the arithmetic mean of Phase0, 1 and 2 Davg. Correlation between V reduction and Davg variation was investigated. Results: 6 patients have been enrolled until now. IMRT treatment prescriptions ranged from 54 to 70Gy. V reduction was seen in all the parotids included, varying from 16,3 to 56,0 % (mean 36,5). ΔD was positive for 5 parotids and negative for 7 parotids; for the subgroup of parotids for which Davg < 20 Gy, average ΔD was negative, while for non-preserved parotids (Davg > 50 Gy) average ΔD was positive. For parotids with Davg closer to the 26 Gy threshold, the behaviour of ΔD was ambiguous, with both massive increments (+ 464 cGy starting from 2759 cGy) and decrements (- 427cGy starting from 4571 cGy). There were no correlation between Davg and ΔD (R2= 0,151); lack of correlation was also showed between V and ΔD (R2= 0,148). Patients’ enrolling is still undergoing. Conclusions: Preservation of parotids for SCCHN patients treated with IMRT has to be pursued keeping into account the unpredictable and drastic variation in Davg attributed to the gland. V variation and initial Davgare not good predictors of ΔD.
POSTER: PHYSICS OPTIMISATION
TRACK:
TREATMENT
PLAN
PO-0848 Towards automated treatment planning in radiotherapy S. Breedveld1 1 Erasmus Medical Center Rotterdam, Radiation Oncology, Rotterdam, The Netherlands
compare different coplanar and non-coplanar beam set-ups for conventional linac and CyberKnife. The lack of posterior beams in the CyberKnife system did not significantly influence the achieved plan quality. In another study, adaptive strategies in liver SBRT were investigated. It was demonstrated that automated daily re-optimization of fluence profiles did almost as well as re-optimization of both beam profiles and beam angles. iCycle is now in daily routine clinical use for an expanding number of treatment sites. Conclusions: With iCycle it is possible to automatically generate treatment plans. Sofar, all studies and clinical experience point at a quality that is superior to plans made by dosimetrists and physicists. Because of the deterministic nature of the algorithm, iCycle can be used to objectively compare treatment strategies. PO-0849 A method to estimate sweeping window arc therapy (SWAT) treatment plan robustness J. Fleckenstein1, J. Hesser1, F. Wenz1, F. Lohr1 1 University Medical Center Mannheim, Department for Radiation Oncology and Radiotherapy, Mannheim, Germany Purpose/Objective: In VMAT dose deliveries with the SWAT technique, in contrast to conventional step-and-shoot or dynamic IMRT with discrete gantry angles, the effect of intrafractional motion on the dose distribution needs further evaluation. With a fast MLC and flattening filter free (FFF) dose delivery, treatment times close to 60 s per fraction are clinical reality. For these treatment times the human breathing period gets close to the period of the collimator sweeping period. Therefore methods to a) quantify the effect of intrafractional target motion and b) provide a sensitive parameter during treatment planning to avoid potentially non-robust treatment plans are presented. Materials and Methods: An Elekta Synergy LINAC with MLCi2 MLC and an Elekta VersaHD LINAC with Agility MLC were used for deliveries of 50 clinical treatment sequences. All treatment deliveries were performed with one 360° VMAT arc and were generated with the Elekta Monaco (V. 3.2-3.3) treatment planning system. Measurements were performed with an ionization chamber array (MatriXX Evolution, IBA) inside a Multicube Lite phantom (IBA) with a sampling time of 100 ms. To detect interplay effects along all three coordinate axes, measurements were performed twice with the detector aligned in vertical or horizontal direction. All slices were moved against each other in MATLAB according to a pre-defined motion vector, mimicking a cosine-to-the-square motion. The minimum relative equivalent uniform dose (EUD) for all breathing starting phases Φ was determined for every amplitude A and period T.
Purpose/Objective: Treatment plan optimization is a labour-intensive time-consuming task, with quality highly dependent on the skills and experience of the planner. Combined with the multi-objective origin of the problem and large degrees of freedom in treatment device setup, the resulting treatment plan may be suboptimal. To overcome these limitations, we developed an algorithm, iCycle, to fully automate treatment planning.
Furthermore LINAC logfiles were recorded during plan delivery. The MLC, jaw, gantry angle and monitor unit settings were continuously saved and used to calculate the cross correlation coefficient between the target motion and the collimator motion component for each direction (CC, LR, AP) separately. The cross correlation coefficient ρ of the dose weighed collimator positions xc and the breathing positions xt was expressed as a product of the standard scores as:
Materials and Methods: To automate the decision making process in treatment planning, iCycle uses lexicographic optimization where hard constraints and prioritized treatment objectives are summarized in a socalled wish-list. Constraints are met unconditionally, while the objectives are optimized in priority. The focus lies on attaining the treatment goals (in priority), where a Pareto optimal plan is obtained after maximally optimizing all objectives. iCycle has an advanced option for coplanar and non-coplanar beam angle selection, integrated with IMRT beam profile optimization. A wish-list is tuned in a collaboration between clinicians and dosimetrists, to be used for a group of similar patients, e.g. all prostate patients with radical treatment. There is no need for adaptation for individual patients. Due to the automation, iCycle is highly suited for objectively comparing treatment strategies, based on large numbers of automatically generated treatment plans.
Results: Figure 1 shows the minimum relative EUD (upper figure) in comparison to the cross correlation coefficient (lower figure) as functions of the period. In the presented case motion in left-right direction was applied. Plan parameters were: Treatment time 102 s, Agility MLC, 20°increment, 179 control points, 2.0 Gy fraction dose, 655.7 MU. The minimum relative EUD observed in all cases was 94.6%. For all cases a cross correlation coefficient above 0.5 corresponded to a minimum in EUD.
Results: iCycle was used for several studies. In a prospective clinical head-and-neck study, both manually generated plans and plans generated by iCycle were blindly presented to the treating clinician. In 32/33 cases, the treating clinician selected the plan generated by iCycle. A treatment planning study for prostate cancer patients with one or two metal hip prosthesis resulted in recommandations for complex cases. For prostate SBRT, iCycle generated 1500 plans automatically to