Efficient and accurate commissioning and quality assurance of radiosurgery beam via prior-embedded implicit neural representation learning

Abstract

Background: Dosimetric commissioning and quality assurance (QA) for linear accelerators (LINACs) present a significant challenge for clinical physicists due to the high measurement workload and stringent precision standards. This challenge is exacerbated for radiosurgery LINACs because of increased measurement uncertainty and more demanding setup accuracy for small-field beams. Optimizing physicists' effort during beam measurements while ensuring the quality of the measured data is crucial for clinical efficiency and patient safety.

Purpose: To develop a radiosurgery LINAC beam model that embeds prior knowledge of beam data through implicit neural representation (NeRP) learning and to evaluate the model's effectiveness in guiding beam data sampling, predicting complete beam dataset from sparse samples, and verifying detector choice and setup during commissioning and QA.

Materials and methods: Beam data including lateral profile and tissue-phantom-ratio (TPR), collected from CyberKnife LINACs, were investigated. Multi-layer perceptron (MLP) neural networks were optimized to parameterize a continuous function of the beam data, implicitly defined by the mapping from measurement coordinates to measured dose values. Beam priors were embedded into network weights by first training the network to learn the NeRP of a vendor-provided reference dataset. The prior-embedded network was further fine-tuned with sparse clinical measurements and used to predict unacquired beam data. Prospective and retrospective evaluations of different beam data samples in finetuning the model were performed using the reference beam dataset and clinical testing datasets, respectively. Model prediction accuracy was evaluated over 10 clinical datasets collected from various LINACs with different manufacturing modes and collimation systems. Model sensitivity in detecting beam data acquisition errors including inaccurate detector positioning and inappropriate detector choice was evaluated using two additional datasets with intentionally introduced erroneous samples.

Results: Prospective and retrospective evaluations identified consistent beam data samples that are most effective in fine-tuning the model for complete beam data prediction. Despite of discrepancies between clinical beam and the reference beam, fine-tuning the model with sparse beam profile measured at a single depth or with beam TPR measured at a single collimator size predicted beam data that closely match ground truth water tank measurements. Across the 10 clinical beam datasets, the averaged mean absolute error (MAE) in percentage dose was lower than 0.5% and the averaged 1D Gamma passing rate (1%/0.5 mm for profile and 1%/1 mm for TPR) was higher than 99%. In contrast, the MAE and Gamma passing rates were above 1% and below 95% between the reference beam dataset and clinical beam datasets. Model sensitivity to beam data acquisition errors was demonstrated by significant model prediction changes when fine-tuned with erroneous versus correct beam data samples, as quantified by a Gamma passing rate as low as 18.16% between model predictions.

Conclusion: A model for small-field radiosurgery beam was proposed that embeds prior knowledge of beam properties and predicts the entire beam data from sparse measurements. The model can serve as a valuable tool for clinical physicists to verify the accuracy of beam data acquisition and promises to improve commissioning and QA reliability and efficiency with substantially reduced number of beam measurements.

Keywords: beam modeling; machine learning; small field dosimetry.

FIGURE 1

General workflow of prior‐embedded NeRP learning for radiosurgery beam modeling.

FIGURE 2

Beam datasets acquired with inappropriate detector choice/inaccurate detector positioning. The first beam dataset consists of beam profile data measured using a small volume ion chamber detector. The detector is suitable for large cone sizes (a) and shows good consistency with diode detector measurements. However, the detector is not appropriate for small cone sizes (b) and shows discrepancies from diode detector measurements. The second beam dataset consists of TPR measurements where TPR at cone sizes less than 25 mm were acquired with a 5 mm shift in depth (c).

FIGURE 3

Model‐based beam data acquisition error detection. The prior‐embedded network was finetuned with different beam data samples and predicted different full beam datasets. Pairwise comparison across network predictions was performed to evaluate the change of model performance when the measured beam dataset contains errors, as one potential use of the model in verifying beam data acquisition accuracy.

FIGURE 4

Beam profile with a 40 mm cone measured at 100 mm depth from a clinical CyberKnifeTM LINAC, overlaid with vendor‐provided reference (a), model predictions when finetuned using beam profile samples measured at 300 mm depth (b) and 15 mm depth (c). Green arrows point to degraded model accuracy when finetuning the model with beam profile measured at 15 mm depth instead of 300 mm depth.

FIGURE 5

Beam TPR measured from a clinical CyberKnifeTM LINAC with a 5 mm cone, overlaid with vendor‐provided reference (a), model predictions when finetuned using beam TPR sample measured with a 20 mm cone (b) and 60 mm cone (c). Green arrows point to degraded model accuracy when finetuning the model with beam TPR measured with the 60 mm cone instead of the 20 mm cone.

FIGURE 6

Prospective and retrospective evaluations of different beam data samples in predicting complete beam TPR (a) and profile (b) datasets. The optimal data sample identified by the prospective method (the one that results in the most accuracy degradation/lowest Gamma passing rates for reference beam dataset modeling) agrees with the optimal sample identified by the retrospective method (the one that results in the highest accuracy/Gamma passing rates in predicting testing clinical beam datasets).

FIGURE 7

Example model‐based self‐consistency check results of beam profile datasets. (a) Beam profile (15 mm cone) predicted by 3 networks finetuned with beam data samples acquired using an ion chamber detector. Net 1 finetuned with small cone (5 mm) beam data sample predicted beam data inconsistently against net 2 and net 3, which were finetuned with large cone (50 mm and 60 mm) beam data samples. (b) Pairwise Gamma passing rates were below 70% when comparing predictions by different networks finetuned with beam data samples acquired using the ion chamber detector. (c) Pairwise Gamma passing rates were all above 95% when comparing predictions by networks finetuned with beam data samples acquired from the same LINAC using a diode detector.

FIGURE 8

Example model‐based self‐consistency check results of a beam TPR dataset acquired with detector position shifts during a subset of measurements. (a) Net 1 finetuned with beam data sample acquired with detector position shift predicts beam data (TPR at 7.5 mm cone) inconsistently against net 2 and net 3, which were finetuned with beam data samples acquired without detector position shifts. (b) Pairwise Gamma passing rates were below 20% when applying the model‐based self‐consistency check scheme to the beam TPR dataset, suggesting the beam data may have been acquired with inconsistent measurement setups.