![\"AI](\"https:\/\/rtmedical.com.br\/wp-content\/uploads\/2026\/06\/ai-dose-guardrails.jpg\")
This article examines these questions from the perspective of medical physicists, dosimetrists, and radiation oncologists who need to make adoption or oversight decisions for AI-based tools. The text differentiates physical description of the phenomenon, commercial implementation and published validation evidence \u2014 three dimensions often confused in discussions on the topic. It is not intended to recommend specific products, but to provide a conceptual map for critical evaluation of these technologies.<\/p>\n
In this Article<\/h2>\n\n- 1. What it means to use AI as a dose surrogate model<\/a><\/li>\n
- 2. Difference between predicting dose and transporting particles<\/a><\/li>\n
- 3. Training Data, Bias, and Validity Domain<\/a><\/li>\n
- 4. Generalization to machines, energies and anatomies<\/a><\/li>\n
- 5. Uncertainty, outlier detection and silent failures<\/a><\/li>\n
- 6. How to compare AI, Monte Carlo and deterministic solvers<\/a><\/li>\n
- 7. Clinical validation, governance and responsible use<\/a><\/li>\n
- 8. FAQ<\/a><\/li>\n
- 9. References<\/a><\/li>\n<\/ul>\n<\/div>\n
What it means to use AI as a dose surrogate model<\/h2>\n
A surrogate model (surrogate model<\/em> or emulator<\/em>) is a computational system trained to reproduce the behavior of another more expensive or slower system, accepting the same inputs and producing approximate outputs. In the dose context, the “expensive” system is typically a high-fidelity MC engine or a linear Boltzmann transport equation (LBTE) solver such as Acuros XB. The surrogate model \u2014 typically a deep convolutional neural network, often with a similar architecture to U-Net \u2014 learns, from reference (input, reference output) pairs, a mapping that can be evaluated in milliseconds rather than minutes or hours.<\/p>\nIt is important to distinguish two sub-cases that the literature often conflicts with. In the first, the network predicts the dose based on already optimized treatment plans<\/em>, functioning as a quick check or generation of an initial plan (knowledge-based planning<\/em>). In the second, the network directly replaces the calculation engine within TPS (treatment planning system<\/em>), being invoked during each optimization iteration. The second case imposes much more severe requirements on accuracy and robustness: a systematically low error in a critical region will propagate to the optimization, producing plans with lower actual coverage than projected, without any warning signal to the user.<\/p>\nThere are prototypes and products that use machine learning in the planning and dose estimation stages, but the intended use must be verified in the documentation for each version. The distinction between “AI-accelerated” and “MC\/LBTE-calculated with hardware acceleration” is crucial. GPUMCD, for example, is Monte Carlo on GPU, not a neural network.<\/p>\n
Reduced latency can support adaptive flows and repeated calculations. The cost is to transfer part of the performance guarantee to the data, validity domain, and failure detection controls.<\/p>\n
Difference between predicting dose and transporting particles<\/h2>\n
Transporting particles, in the physical sense, means solving \u2014 exactly, approximately or stochastically \u2014 the Boltzmann equation for radiation transport, considering interaction cross sections dependent on the material crossed, locally deposited energy and secondary scattering. MC samples individual trajectories of photons, electrons and secondary particles. LBTE\/Acuros XB solves the equation in its deterministic form over a spatial mesh. Pencil Beam decomposes the beam into pencils and applies water-calibrated scattering kernels, with empirical corrections for inhomogeneities. The AAA (Anisotropic Analytical Algorithm<\/em>) utilizes separate energy convolutions for primary photons, lateral scattering, and contaminating electrons. All of these algorithms have parameters with direct physical meaning and can be, at least in principle, commissioned and validated against independent phantom measurements.<\/p>\nA dose prediction neural network does not solve any of these equations. It learns a function \u2014 potentially of very high dimensionality \u2014 that maps the geometry of the problem (CT morphology in Hounsfield units, structure contours, beam configuration) to a dose distribution, minimizing a loss functional over the training set. The learned mapping is, by construction, an interpolation over the manifold of cases seen during training. Outside of this manifold \u2014 an unusual anatomy, an unrepresented combination of energies, an atypical beam geometry \u2014 the network will extrapolate in an unpredictable manner, with no guarantee of physical coherence.<\/p>\n
This distinction has direct implications for concepts such as dose to medium<\/em> (Dm) and dose to water<\/em> (Dw). Algorithms such as Acuros XB allow you to explicitly choose which quantity is calculated, with clinical consequences discussed in the literature especially in bone-tissue interfaces and in proton therapy. A surrogate model trained on Dm outputs implicitly “learns” this convention, but will not make it explicit. A convention change in the reference engine during retraining may go unnoticed \u2014 a structural example of silent failure.<\/p>\nAnother relevant aspect is incremental convergence: in MC, more particle histories equate to lower statistical uncertainty, and the user can balance calculation time and accuracy in a controlled way. In an ML model, there is no equivalent mechanism \u2014 the output is deterministic for a given input, and the model’s uncertainty is fixed, determined by the training phase.<\/p>\n
Training Data, Bias, and Validity Domain<\/h2>\n
The performance of any surrogate model is fundamentally limited by the quality, quantity, and diversity of the training data. For dose prediction, the data set is generally clinically approved plans at one or more institutions, with dose distributions calculated by institutional TPS as the label (ground truth<\/em>). Two structural problems immediately emerge.<\/p>\nFirst, the label is not the actual dose \u2014 it is the dose calculated by the TPS algorithm, with its own uncertainties and approximations. If TPS used Pencil Beam for lung cases with severe heterogeneities, and the model learns to reproduce Pencil Beam, there is no gain in physical accuracy; there is only acceleration of an imprecise method. Second, the training data reflects local planning patterns and biases: preferred beam topologies, normalization criteria, margin philosophies. A model trained in a highly specialized center may not generalize to a center with different patient populations, equipment or practices.<\/p>\n
The table below summarizes the most relevant sources of bias in training datasets for dose models:<\/p>\n
What it means to use AI as a dose surrogate model<\/h2>\n
A surrogate model (surrogate model<\/em> or emulator<\/em>) is a computational system trained to reproduce the behavior of another more expensive or slower system, accepting the same inputs and producing approximate outputs. In the dose context, the “expensive” system is typically a high-fidelity MC engine or a linear Boltzmann transport equation (LBTE) solver such as Acuros XB. The surrogate model \u2014 typically a deep convolutional neural network, often with a similar architecture to U-Net \u2014 learns, from reference (input, reference output) pairs, a mapping that can be evaluated in milliseconds rather than minutes or hours.<\/p>\n It is important to distinguish two sub-cases that the literature often conflicts with. In the first, the network predicts the dose based on already optimized treatment plans<\/em>, functioning as a quick check or generation of an initial plan (knowledge-based planning<\/em>). In the second, the network directly replaces the calculation engine within TPS (treatment planning system<\/em>), being invoked during each optimization iteration. The second case imposes much more severe requirements on accuracy and robustness: a systematically low error in a critical region will propagate to the optimization, producing plans with lower actual coverage than projected, without any warning signal to the user.<\/p>\n There are prototypes and products that use machine learning in the planning and dose estimation stages, but the intended use must be verified in the documentation for each version. The distinction between “AI-accelerated” and “MC\/LBTE-calculated with hardware acceleration” is crucial. GPUMCD, for example, is Monte Carlo on GPU, not a neural network.<\/p>\n Reduced latency can support adaptive flows and repeated calculations. The cost is to transfer part of the performance guarantee to the data, validity domain, and failure detection controls.<\/p>\n Transporting particles, in the physical sense, means solving \u2014 exactly, approximately or stochastically \u2014 the Boltzmann equation for radiation transport, considering interaction cross sections dependent on the material crossed, locally deposited energy and secondary scattering. MC samples individual trajectories of photons, electrons and secondary particles. LBTE\/Acuros XB solves the equation in its deterministic form over a spatial mesh. Pencil Beam decomposes the beam into pencils and applies water-calibrated scattering kernels, with empirical corrections for inhomogeneities. The AAA (Anisotropic Analytical Algorithm<\/em>) utilizes separate energy convolutions for primary photons, lateral scattering, and contaminating electrons. All of these algorithms have parameters with direct physical meaning and can be, at least in principle, commissioned and validated against independent phantom measurements.<\/p>\n A dose prediction neural network does not solve any of these equations. It learns a function \u2014 potentially of very high dimensionality \u2014 that maps the geometry of the problem (CT morphology in Hounsfield units, structure contours, beam configuration) to a dose distribution, minimizing a loss functional over the training set. The learned mapping is, by construction, an interpolation over the manifold of cases seen during training. Outside of this manifold \u2014 an unusual anatomy, an unrepresented combination of energies, an atypical beam geometry \u2014 the network will extrapolate in an unpredictable manner, with no guarantee of physical coherence.<\/p>\n This distinction has direct implications for concepts such as dose to medium<\/em> (Dm) and dose to water<\/em> (Dw). Algorithms such as Acuros XB allow you to explicitly choose which quantity is calculated, with clinical consequences discussed in the literature especially in bone-tissue interfaces and in proton therapy. A surrogate model trained on Dm outputs implicitly “learns” this convention, but will not make it explicit. A convention change in the reference engine during retraining may go unnoticed \u2014 a structural example of silent failure.<\/p>\n Another relevant aspect is incremental convergence: in MC, more particle histories equate to lower statistical uncertainty, and the user can balance calculation time and accuracy in a controlled way. In an ML model, there is no equivalent mechanism \u2014 the output is deterministic for a given input, and the model’s uncertainty is fixed, determined by the training phase.<\/p>\n The performance of any surrogate model is fundamentally limited by the quality, quantity, and diversity of the training data. For dose prediction, the data set is generally clinically approved plans at one or more institutions, with dose distributions calculated by institutional TPS as the label (ground truth<\/em>). Two structural problems immediately emerge.<\/p>\n First, the label is not the actual dose \u2014 it is the dose calculated by the TPS algorithm, with its own uncertainties and approximations. If TPS used Pencil Beam for lung cases with severe heterogeneities, and the model learns to reproduce Pencil Beam, there is no gain in physical accuracy; there is only acceleration of an imprecise method. Second, the training data reflects local planning patterns and biases: preferred beam topologies, normalization criteria, margin philosophies. A model trained in a highly specialized center may not generalize to a center with different patient populations, equipment or practices.<\/p>\n The table below summarizes the most relevant sources of bias in training datasets for dose models:<\/p>\nDifference between predicting dose and transporting particles<\/h2>\n
Training Data, Bias, and Validity Domain<\/h2>\n