DoseGAN: a generative adversarial network for synthetic dose prediction using attention-gated discrimination and generation

Abstract

Deep learning algorithms have recently been developed that utilize patient anatomy and raw imaging information to predict radiation dose, as a means to increase treatment planning efficiency and improve radiotherapy plan quality. Current state-of-the-art techniques rely on convolutional neural networks (CNNs) that use pixel-to-pixel loss to update network parameters. However, stereotactic body radiotherapy (SBRT) dose is often heterogeneous, making it difficult to model using pixel-level loss. Generative adversarial networks (GANs) utilize adversarial learning that incorporates image-level loss and is better suited to learn from heterogeneous labels. However, GANs are difficult to train and rely on compromised architectures to facilitate convergence. This study suggests an attention-gated generative adversarial network (DoseGAN) to improve learning, increase model complexity, and reduce network redundancy by focusing on relevant anatomy. DoseGAN was compared to alternative state-of-the-art dose prediction algorithms using heterogeneity index, conformity index, and various dosimetric parameters. All algorithms were trained, validated, and tested using 141 prostate SBRT patients. DoseGAN was able to predict more realistic volumetric dosimetry compared to all other algorithms and achieved statistically significant improvement compared to all alternative algorithms for the V 100 and V 120 of the PTV, V 60 of the rectum, and heterogeneity index. Advanced treatment techniques such as intensity modulated radiation therapy (IMRT) and volumetrically modulated arc therapy (VMAT) have become standard of care for many treatment sites 1,2. Creating clinically acceptable treatment plans using these advanced techniques requires extensive domain expertise and is exceedingly time consuming 3,4. To reduce the burden on clinical resources, the development of automated treatment planning technologies has accelerated in recent years 5-10. Historically, automated treatment planning technologies relied on selecting handcrafted features, such as spatial relationships between planning volumes, overlapping volume histograms, planning volume shapes, planning volume and field intersections, field shapes, planning volume depths , and distance-to-target histograms (DTH) 11-14. These techniques rely on machine learning algorithms such as gradient boosting, random forests, and support vector machines to find strong correlations between groups of weakly correlated predictive features 6,15-17. Such techniques achieve good performance on inherently structured data, but tend to struggle if the problem does not easily reduce to a structured format. Because of this, deep learning approaches have emerged that predict dose using fully connected layers 18. However, fully connected layers tend to not generalize well on highly dimensional data. Convolutional neural networks (CNNs) have emerged to solve many image processing tasks 4,6,19-23. Recently, encoder-decoder CNNs have been used to predict radiation dose from arbitrary patient anatomy. These method...

Artificial intelligence and machine learning for medical imaging: A technology review

“…inaccurate aligning or labelling) [39,97,194,197]. Besides image translation, GANs have also been applied to other tasks, such as segmentation [198][199][200][201][202][203] or radiotherapy dose prediction [204][205][206][207], or artifact reduction [208], among others [183].…”

Artificial intelligence applications in intensity modulated radiation treatment planning: an overview

“…While it is possible to estimate dose distribution using algorithms such as deformable image registration (50), most recent dose prediction algorithms are based on ML or DL models (51). Categorized by prediction algorithms, dose distribution could be predicted by shallow ML models (49,51), deep neural network models such as the convolutional neural network (CNN) (47,(52)(53)(54)(55)(56)(57)(58)(59) and the generative adversarial network (GAN) (48,60,61). Categorized by input/output dimensions, dose distribution could be predicted voxel by voxel (49,51), slice by slice (47,48,52,56,59,62), or as a 3D volume (53,54,57,58,60,61).…”

“…Categorized by prediction algorithms, dose distribution could be predicted by shallow ML models (49,51), deep neural network models such as the convolutional neural network (CNN) (47,(52)(53)(54)(55)(56)(57)(58)(59) and the generative adversarial network (GAN) (48,60,61). Categorized by input/output dimensions, dose distribution could be predicted voxel by voxel (49,51), slice by slice (47,48,52,56,59,62), or as a 3D volume (53,54,57,58,60,61). In the following section, we present one example of a DL based dose prediction model for prostate VMAT.…”

Technical Note: Dose prediction for radiation therapy using feature‐based losses and One Cycle Learning

“…Recent dose prediction models have utilized generative adversarial networks (GANs) to improve conversion performance. 1,3 However, despite the fact that GANs resulted in great success for many different applications, the technique still suffers from shortcomings like vanishing gradients and mode collapse. This makes GANs difficult to train and requires an extensive hyperparameter search for optimal performance.…”