Qian, Jing

Abstract Objective . Phase gating is a critical technique to mitigate tumor motion during radiotherapy, particularly in spot-scanned particle therapy where internal motion can interfere with dynamic spot scanning patterns and, simultaneously, introducing substantial range uncertainties. However, the current commercial state of the art in real-time phase prediction is challenged by patient-specific breathing variability as well as detection and delivery system latencies. This leads to suboptimal efficiency, mis-timed radiation delivery and requires frequent manual intervention. This study aims to improve phase prediction accuracy using deep learning (DL)-based time series forecasting to enable more accurate dose delivery. Approach . Retrospective breathing waveforms from 69 proton therapy patients, sampled at 30 Hz, were labeled with inspiratory peaks and assigned subjective regularity scores (four levels). DL models with various architectures were trained using waveform amplitude to predict current or future breathing phases. Model performance was evaluated using mean squared error, phase binning accuracy, and timing deviation for radiation on/off events. Bayesian optimization was used for hyperparameter tuning. Results were compared between models and against a commercial algorithm currently in clinical use. Main Results . The curated waveform dataset included 165 242 s for training, 24 057 s for validation, and 30 322 s for testing, with an additional 40 604 s from separate patients for extended validation. The long short-term memory and temporal fusion transformer models significantly outperformed the commercial algorithm, improving phase prediction accuracy by nearly 20% and reducing timing deviations across all regularity levels. Significance . DL-based time series forecasting may substantially improve breathing phase prediction accuracy over current clinically available methods, offering a more precise and reliable approach to phase-gated radiation delivery.

Background: Post-treatment prognosis and monitoring are critical for determining the timing of salvage treatment in glioblastoma patients but has been challenging due to difficulties differentiating progression from treatment effects in conventional images. This exploratory study aimed to establish the correlation of radiomics image features from time series of amino acid tracer 18F-DOPA PET images, with outcomes, using machine learning and dimension reduction analysis. Methods: 18F-DOPA PET images were collected for a patient cohort with wild-type IDH and unmethylated MGMT who underwent dose-escalated radiation therapy. Quantitative features were derived from the high uptake region (T/N > 2.0) in pre- and post-radiation therapy follow-up 18F-DOPA PET images. A customized workflow was utilized for pre-selecting predictive features, followed by manifold learning. Machine learning algorithms were employed to establish associations between imaging features and remaining survival (RS), defined as the time between a follow-up scan and date of death. Results: The ML models exhibited 81–83% ROC_AUC in predicting RS evaluated on an independent test dataset. A RS map is proposed for monitoring tumor alterations through serial 18F-DOPA PET scans, demonstrating superior sensitivity and better correlation with survival compared to the RANO criteria. Conclusions: Our study demonstrates that ML models utilizing FU 18F-DOPA PET images have the potential to effectively predict future survival outcomes in patients with glioblastoma treated with dose-escalated radiation therapy. The capability to assess changes in tumor over time through imaging can potentially assist in patient stratification and the selection of salvage treatments, while also aiding in distinguishing treatment effects from genuine tumor progression.

AbstractPurposeThe Leksell Gamma Knife Icon unit (IU) was introduced recently as an upgrade to the Perfexion unit (PU) at our Gamma Knife practice. In the current study, we sought mainly to characterize dosimetry and targeting accuracy of the IU treatment deliveries using both invasive frame and frameless treatment workflows.MethodsRelative output factors were measured by delivering single‐shot 4, 8 and 16 mm radiation profiles in the manufacturer's acrylonitrile butadiene styrene spherical phantom in coronal and sagittal planes using EBT3 film. Resultant dosimetry was compared with the manufacturer's dose calculation and derived output factors were compared with the manufacturer's published value. Geometric consistency of stereotactic coordinates based on cone‐beam computed tomography (CBCT) versus the traditional conventional CT‐based method was characterized using a rigid phantom containing nine fiducial indicators over four separate trials. End‐to‐end (E2E) testing using EBT3 film was designed to evaluate both dosimetric and geometric accuracy for hypothetical framed and frameless workflows.ResultsRelative output factors as measured by the manufacturer were independently confirmed using EBT3 film measurements to within 2%. The mean 3D radial discrepancy in stereotactic space between CBCT and CT‐based definition over the sampled locations in our rigid geometry phantom was demonstrated to be between 0.40 mm and 0.56 mm over the set of trials, larger than prior reported values. E2E performed in 2D demonstrates sub‐mm (and typically < 0.5 mm) accuracy for framed and frameless workflows; geometric accuracy of framed treatments using CBCT‐defined stereotactic coordinates was shown to be slightly improved in comparison with those defined using conventional CT. Furthermore, in phantom, frameless workflows exhibited better accuracy than framed workflows for fractionated treatments, despite large magnitudes of introduced interfraction setup error. Accuracy of dosimetric delivery was confirmed in terms of qualitative comparisons of dose profiles and in terms of 2D gamma pass rates based on 1%/1 mm criteria.ConclusionThe IU was commissioned for clinical use of frameless and framed treatment protocols. The present study outlines an extensive E2E methodology for confirmation of dosimetric and geometric treatment accuracy.