Implicit neural representation for radiation therapy dose distribution

Abstract

Objective. Dose distribution data plays a pivotal role in radiotherapy treatment planning. The data is typically represented using voxel grids, and its size ranges from 10^6--10^8. A concise representation of the treatment plan is of great value in facilitating treatment planning and downstream applications. This work aims to develop an implicit neural representation of 3D dose distribution data. Approach. Instead of storing the dose values at each voxel, in the proposed approach, the weights of a multilayer perceptron (MLP) are employed to characterize the dosimetric data for plan representation and subsequent applications. We train a coordinate-based MLP with sinusoidal activations to map the voxel spatial coordinates to the corresponding dose values. We identify the best architecture for a given parameter budget and use that to train a model for each patient. The trained MLP is evaluated at each voxel location to reconstruct the dose distribution. We perform extensive experiments on dose distributions of prostate, spine, and head and neck tumor cases to evaluate the quality of the proposed representation. We also study the change in representation quality by varying model size and activation function. Main Results. Using coordinate-based MLPs with sinusoidal activations, we can learn implicit representations that achieve a mean-squared error of 10^{-6} and peak signal-to-noise ratio greater than 50 dB at a target bitrate of ~1 across all the datasets, with a compression ratio of ~32. Our results also show that model sizes with a bitrate of 1--2 achieve optimal accuracy. For smaller bitrates, performance starts to drop significantly. Significance. The proposed model provides a low-dimensional, implicit, and continuous representation of 3D dose data. In summary, given a dose distribution, we systematically show how to find a compact model to fit the data accurately. This study lays the groundwork for future applications of neural representations of dose data in radiation oncology.

Sub‐second photon dose prediction via transformer neural networks

“…iDoTA is trained using a certain resolution and grid dimensions, which must be fixed for both training and inference. For dose prediction in finer grid resolutions, iDoTA can be coupled to neural representation models capable of accurate super‐resolution 50 . Regarding grid size, predicting dose distributions from treatment plans or beams through anatomies larger than the predetermined voxel grid must be done in steps, obtaining several input volumes and accumulating the outputs along the beam depth.…”

“…For dose prediction in finer grid resolutions, iDoTA can be coupled to neural representation models capable of accurate super-resolution. 50 Regarding grid size, predicting dose distributions from treatment plans or beams through anatomies larger than the predetermined voxel grid must be done in steps, obtaining several input volumes and accumulating the outputs along the beam depth. Conversely, all doses can be predicted for the same fixed grid covering the part of the anatomy containing the structures of interests, which neglects the (usually) low doses near patient entrance.…”

Sub-second photon dose prediction via transformer neural networks

“…iDoTA is trained using a certain resolution and grid dimensions, which must be fixed for both training and inference. For dose prediction in finer grid resolutions, iDoTA can be coupled to neural representation models capable of accurate super-resolution [49]. Regarding grid size, predicting dose distributions from treatment plans or beams through anatomies larger than the predetermined voxel grid must be done in steps, obtaining several input volumes and accumulating the outputs along the beam depth.…”

Modeling linear accelerator (Linac) beam data by implicit neural representation learning for commissioning and quality assurance applications

“…Similar with previous work on NeRP learning for imaging and treatment planning, 23,24 a MLP network 20,21 was implemented. The network consists of eight fullyconnected layers.…”

“…Recent research has shown significant promise in using NeRP to accurately model traditional discrete representations that use pixel/voxel grids. This technique has been successfully applied to many tasks such as computed tomography (CT) and magnetic resonance image (MRI) reconstruction, 23 dosimetric plan representation, 24 generative modeling, 25 synthesizing high-resolution 3D shapes using coarse voxels, 26,27 compressing 2D images and videos 28 and neural rearrangement planning for unknown objects. 29 By learning the NeRP of the "golden" beam dataset, we assumed that the prior knowledge of the Linac beam data was embedded through network weights and can be used to assist inference of full beam dataset from sparse measurements, where the prior-embedded network was further trained to fit a subset of beam data measured and the complete set of beam data was predicted by querying the trained network.…”