Dose calculation in proton therapy using a discovery cross‐domain generative adversarial network (DiscoGAN)

Abstract

Accurate dose calculation is a critical step in proton therapy. A novel machine learningbased approach was proposed to achieve comparable accuracy to that of Monte Carlo simulation while reducing the computational time. Methods: Computed tomography-based patient phantoms were used and three treatment sites were selected (thorax, head, and abdomen), comprising different beam pathways and beam energies. The training data were generated using Monte Carlo simulations. A discovery cross-domain generative adversarial network (DiscoGAN) was developed to perform the mapping between two domains: stopping power and dose, with HU values from CT images incorporated as auxiliary features. The accuracy of dose calculation was quantitatively evaluated in terms of mean relative error (MRE) and mean absolute error (MAE). The relationship between the DiscoGAN performance and other factors such as absolute dose, beam energy and location within the beam cross-section (center and off-center lines) was examined. Results: The DiscoGAN model is found to be effective in dose calculation. For the abdominal case, the MRE is found to 1.47% (mean), 3.30% (maximum) and 0.67% (minimum). For the thoracic case, the MRE is found to~2.43% (mean), 4.80% (maximum) and 0.71% (minimum). For the head case, the MRE is found to~2.83% (mean), 4.84% (maximum) and 1.01% (minimum). Comparable accuracy is found in the independent validation dataset (different CT images), achieving a mean MRE of 1.65% (thorax), 4.02% (head) and 1.64% (abdomen). For the energy span between 80 and 130 MeV, no strong dependency of accuracy on beam energy is found. The results imply that no systematic deviation, either over-dose or under-dose, occurs between the predicted dose and raw dose. Conclusion:The DiscoGAN framework demonstrates great potential as a tool for dose calculation in proton therapy, achieving comparable accuracy yet being more efficient relative to Monte Carlo simulation. Its comparison with the pencil beam algorithm (PBA) will be the next step of our research. If successful, our proposed approach is expected to find its use in more advanced applications such as inverse planning and adaptive proton therapy.

Feasibility study of fast intensity‐modulated proton therapy dose prediction method using deep neural networks for prostate cancer

“…From the perspective of physics, multiple Coulomb scattering processes are strongly tied to tissue composition (i.e., electron density and cross-section). 21 Neither the analytically derived SPR in our study nor the PB is able to incorporate that as accurately as MC simulation. Therefore, the under-dose or over-dose would occur in heterogeneous regions, making dose calculation less accurate.…”

“…19 In addition to U-net, the recurrent neural network and generative adversarial network are also used to predict the 3D proton dose because it is understood as a sequence of 2D dose slices along the beam direction. 20,21 Since the IMPT plans consist of scanning spots with different energies, the prediction accuracy relies heavily on the plan information and beam data. Nowadays, no suitable model for dose prediction of IMPT plans has been reported yet.…”

“…Guerreiro et al predicted proton and photon dose distributions for pediatric abdominal tumors based on patient‐specific anatomy, allowing fast dose visualization to aid in the selection of effective radiotherapy techniques 19 . In addition to U‐net, the recurrent neural network and generative adversarial network are also used to predict the 3D proton dose because it is understood as a sequence of 2D dose slices along the beam direction 20,21 . Since the IMPT plans consist of scanning spots with different energies, the prediction accuracy relies heavily on the plan information and beam data.…”

Machine learning based oxygen and carbon concentration derivation using dual‐energy CT for PET‐based dose verification in proton therapy

“…As previously mentioned in section 2.5, this suffices our intention to pinpoint the extent of performance degradation due to external perturbation such as elemental composition fluctuation or image noise. For quantitative analysis, mean relative error (MRE) was used to evaluate the difference between the simulated activity and their counterparts in the ground truth, in the same manner as in our previous work 18,19,26 :…”

Direct mapping from PET coincidence data to proton-dose and positron activity using a deep learning approach

“…The second sequence of the RNN model-based study combined the PET detector model, proton beam stopping power, and anatomical information using the reconstructed dose-mapping profile (Hu et al 2020). Another study (Zhang et al 2021) used deep learning methods to predict proton dose in 3D. The study was performed on a ring-shaped PET prototype with 88,000 crystal elements to reconstruct PET activity using single slice re-binning reconstruction and 2D OSEM iterative reconstruction algorithms.…”