Small beams, fast predictions: a comparison of machine learning dose prediction models for proton minibeam therapy

Abstract

Background Dose calculations for novel radiotherapy cancer treatments such as proton minibeam radiation therapy is often done using full Monte Carlo (MC) simulations. As MC simulations can be very time consuming for this kind of application, deep learning models have been considered to accelerate dose estimation in cancer patients. Purpose This work systematically evaluates the dose prediction accuracy, speed and generalization performance of three selected state‐of‐the‐art deep learning models for dose prediction applied to the proton minibeam therapy. The strengths and weaknesses of those models are thoroughly investigated, helping other researchers to decide on a viable algorithm for their own application. Methods The following recently published models are compared: first, a 3D U‐Net model trained as a regression network, second, a 3D U‐Net trained as a generator of a generative adversarial network (GAN) and third, a dose transformer model which interprets the dose prediction as a sequence translation task. These models are trained to emulate the result of MC simulations. The dose depositions of a proton minibeam with a diameter of 800μm and an energy of 20–100 MeV inside a simple head phantom calculated by full Geant4 MC simulations are used as a case study for this comparison. The spatial resolution is 0.5 mm. Special attention is put on the evaluation of the generalization performance of the investigated models. Results Dose predictions with all models are produced in the order of a second on a GPU, the 3D U‐Net models being fastest with an average of 130 ms. An investigated 3D U‐Net regression model is found to show the strongest performance with overall 61.0%±$\%\pm$0.5% of all voxels exhibiting a deviation in energy deposition prediction of less than 3% compared to full MC simulations with no spatial deviation allowed. The 3D U‐Net models are observed to show better generalization performance for target geometry variations, while the transformer‐based model shows better generalization with regard to the proton energy. Conclusions This paper reveals that (1) all studied deep learning models are significantly faster than non‐machine learning approaches predicting the dose in the order of seconds compared to hours for MC, (2) all models provide reasonable accuracy, and (3) the regression‐trained 3D U‐Net provides the most accurate predictions.

Evaluation of an automated clinical decision system with deep learning dose prediction and NTCP model for prostate cancer proton therapy

“…To enhance the availability of the dose prediction model, we chose the 3D UNet over several other dose prediction models due to its performance. Benefiting from the encoder-decoder structure that captures high-level semantics while preserving low-level spatial details, 3D UNet and its variants are widely used for various medical image segmentation (Çiçek et al 2016(Çiçek et al , Fu et al 2023 and dose prediction tasks in radiotherapy (Guerreiro et al 2021, Mashayekhi et al 2022 (Mentzel et al 2022). To investigate the effect of training set size on IMPT and VMAT dose prediction, we trained the 3D UNet using 45, 55, and 65 cases.…”

“…One potential solution to accelerate and standardize the plan comparison process is the use of auto dose prediction. Recently, deep learning techniques have shown potential in predicting dose distribution without manual intervention (Kearney et al 2018, Fan et al 2019, Nomura et al 2020, Guerreiro et al 2021, Neishabouri et al 2021, Mashayekhi et al 2022, Mentzel et al 2022, Wang et al 2022. While numerous studies before 2020 focused on predicting photon dose distributions using Residual Network, three-dimensional convolutional neural network (3D CNN), 3D UNet, and other networks (Kearney et al 2018, Fan et al 2019, limited research existed on predicting proton dose distribution.…”

Accurate and Fast Deep Learning Dose Prediction for a Preclinical Microbeam Radiation Therapy Study Using Low-Statistics Monte Carlo Simulations

“…In recent years, an increasing number of studies investigating fast dose predictions for radiotherapy treatment planning with GPU algorithms [ 1 ] and deep learning models have been published [ 2 , 3 , 4 , 5 ]. However, these publications mostly focus on clinically available treatment methods, such as IMRT [ 6 , 7 ], VMAT [ 8 , 9 ] or proton pencil beam scanning [ 10 ].…”