A machine learning tool for re-planning and adaptive RT: A multicenter cohort investigation

PURPOSE

To predict patients who would benefit from adaptive radiotherapy (ART) and re-planning intervention based on machine learning from anatomical and dosimetric variations in a retrospective dataset.

MATERIALS AND METHODS

90 patients (pts) treated for head-neck cancer (H&N) formed a multicenter data-set. 41 H&N pts (45.6%) were considered for learning; 49 pts (54.4%) were used to test the tool. A homemade machine-learning classifier was developed to analyze volume and dose variations of parotid glands (PG). Using deformable image registration (DIR) and GPU, patients' conditions were analyzed automatically. Support Vector Machines (SVM) was used for time-series evaluation. "Inadequate" class identified patients that might benefit from replanning. Double-blind evaluation by two radiation oncologists (ROs) was carried out to validate day/week selected for re-planning by the classifier.

RESULTS

The cohort was affected by PG mean reduction of 23.7±8.8%. During the first 3weeks, 86.7% cases show PG deformation aligned with predefined tolerance, thus not requiring re-planning. From 4th week, an increased number of pts would potentially benefit from re-planning: a mean of 58% of cases, with an inter-center variability of 8.3%, showed "inadequate" conditions. 11% of cases showed "bias" due to DIR and script failure; 6% showed "warning" output due to potential positioning issues. Comparing re-planning suggested by tool with recommended by ROs, the 4th week seems the most favorable time in 70% cases.

CONCLUSIONS

SVM and decision-making tool was applied to overcome ART challenges. Pts would benefit from ART and ideal time for re-planning intervention was identified in this retrospective analysis.

In silico trial of simulation-free hippocampal-avoidance whole brain adaptive radiotherapy

Background and Purpose

Hippocampal-avoidance whole brain radiotherapy (HA-WBRT) can be a time-consuming process compared to conventional whole brain techniques, thus potentially limiting widespread utilization. Therefore, we evaluated the in silico clinical feasibility, via dose-volume metrics and timing, by leveraging a computed tomography (CT)-based commercial adaptive radiotherapy (ART) platform and workflow in order to create and deliver patient-specific, simulation-free HA-WBRT.

Materials and methods

Ten patients previously treated for central nervous system cancers with cone-beam computed tomography (CBCT) imaging were included in this study. The CBCT was the adaptive image-of-the-day to simulate first fraction on-board imaging. Initial contours defined on the MRI were rigidly matched to the CBCT. Online ART was used to create treatment plans at first fraction. Dose-volume metrics of these simulation-free plans were compared to standard-workflow HA-WBRT plans on each patient CT simulation dataset. Timing data for the adaptive planning sessions were recorded.

Results

For all ten patients, simulation-free HA-WBRT plans were successfully created utilizing the online ART workflow and met all constraints. The median hippocampi D100% was 7.8 Gy (6.6-8.8 Gy) in the adaptive plan vs 8.1 Gy (7.7-8.4 Gy) in the standard workflow plan. All plans required adaptation at first fraction due to both a failing hippocampal constraint (6/10 adaptive fractions) and sub-optimal target coverage (6/10 adaptive fractions). Median time for the adaptive session was 45.2 min (34.0-53.8 min).

Conclusions

Simulation-free HA-WBRT, with commercially available systems, was clinically feasible via plan-quality metrics and timing, in silico.

Validation of a Monte Carlo-based dose calculation engine including the 1.5 T magnetic field for independent dose-check in MRgRT

PURPOSE

Adaptive MRgRT by 1.5 T MR-linac requires independent verification of the plan-of-the-day by the primary TPS (MonacoTM) (M). Here we validated a Monte Carlo-based dose-check including the magnetostatic field, SciMoCaTM (S).

METHODS

M and S were validated first in water, by comparison with commissioning-dosimetry. PDD(2x2cm2) through a lung(air)-equivalent virtual-slab was then calculated. Clinical validation retrospectively included 161 SBRT plans, from five patients per-site: Pelvic-Nodes, Prostate, Liver, Pancreas, and Lungs. S-minus-M percentage differences (Δ%) were computed for target- and OARs-related dose-volume metrics. In-phantom dose verification per-patient was performed.

RESULTS

γ(2 %,1mm)-passing-rates (PR%) of in-water-computed PDD and transverse-dose-profiles vs. commissioning-dosimetry were (99.1 ± 2.0)% for M, and (99.3 ± 1.5)% for S. Calculated output-factors (OF) were typically within 1 % from measurements, except for OF(1x1cm2) which was misestimated by -4.4 % and + 2.2 %, by M and S respectively. Dose spikes (valleys) on the PDD(2x2cm2) by S across the lung-equivalent virtual-slab were slightly reduced with respect to M. In clinical plans, S computed higher V95% (p <0.05*, for pancreas and lung) and D2% (p <0.05*, for all sites) for the target, while D%>2% resulted for duodenal D(1cm3), in Pancreas-SBRT, and for mean-lung-dose, in Lung-SBRT. All mostly due to the underestimated OF(1x1cm2) by M. In-phantom dose verifications showed an average 1% increase in PR% by S vs. M.

CONCLUSIONS

Beam-model quality in S resulted equivalent to M, thus making S useful both for an independent validation of the same beam-model in M, and for a daily validation of the M-based online approval decisions, without significantly delaying the clinical workflow (2-3 min).