Incorporating human and learned domain knowledge into training deep neural networks: A differentiable dose‐volume histogram and adversarial inspired framework for generating Pareto optimal dose distributions in radiation therapy

Abstract

Purpose We propose a novel domain‐specific loss, which is a differentiable loss function based on the dose‐volume histogram (DVH), and combine it with an adversarial loss for the training of deep neural networks. In this study, we trained a neural network for generating Pareto optimal dose distributions, and evaluate the effects of the domain‐specific loss on the model performance. Methods In this study, three loss functions — mean squared error (MSE) loss, DVH loss, and adversarial (ADV) loss — were used to train and compare four instances of the neural network model: (a) MSE, (b) MSE + ADV, (c) MSE + DVH, and (d) MSE + DVH+ADV. The data for 70 prostate patients, including the planning target volume (PTV), and the organs at risk (OAR) were acquired as 96 × 96 × 24 dimension arrays at 5 mm3 voxel size. The dose influence arrays were calculated for 70 prostate patients, using a 7 equidistant coplanar beam setup. Using a scalarized multicriteria optimization for intensity‐modulated radiation therapy, 1200 Pareto surface plans per patient were generated by pseudo‐randomizing the PTV and OAR tradeoff weights. With 70 patients, the total number of plans generated was 84 000 plans. We divided the data into 54 training, 6 validation, and 10 testing patients. Each model was trained for a total of 100,000 iterations, with a batch size of 2. All models used the Adam optimizer, with a learning rate of 1 × 10−3. Results Training for 100 000 iterations took 1.5 days (MSE), 3.5 days (MSE+ADV), 2.3 days (MSE+DVH), and 3.8 days (MSE+DVH+ADV). After training, the prediction time of each model is 0.052 s. Quantitatively, the MSE+DVH+ADV model had the lowest prediction error of 0.038 (conformation), 0.026 (homogeneity), 0.298 (R50), 1.65% (D95), 2.14% (D98), and 2.43% (D99). The MSE model had the worst prediction error of 0.134 (conformation), 0.041 (homogeneity), 0.520 (R50), 3.91% (D95), 4.33% (D98), and 4.60% (D99). For both the mean dose PTV error and the max dose PTV, Body, Bladder and rectum error, the MSE+DVH+ADV outperformed all other models. Regardless of model, all predictions have an average mean and max dose error <2.8% and 4.2%, respectively. Conclusion The MSE+DVH+ADV model performed the best in these categories, illustrating the importance of both human and learned domain knowledge. Expert human domain‐specific knowledge can be the largest driver in the performance improvement, and adversarial learning can be used to further capture nuanced attributes in the data. The real‐time prediction capabilities allow for a physician to quickly navigate the tradeoff space for a patient, and produce a dose distribution as a tangible endpoint for the dosimetrist to use for planning. This is expected to considerably reduce the treatment planning time, allowing for clinicians to focus their efforts on the difficult and demanding cases.

Fully automated dose prediction using generative adversarial networks in prostate cancer patients

“…When we compared our results with these previous results, we found that the average absolute dose differences of most evaluation metrics in the structure-based model were smaller than those of Nguyen et al by using the GAN (Table 4). Some papers demonstrated that the prediction of the GAN outperformed the U-net based architecture [5,11], and this tendency was also seen in the present study (Table 4). Moreover, an extremely small deviation was observed in both prediction models and the prediction performance was comparable to the previous results even when the CT images were used for the training (Table 4).…”

“…In the fairly recent past, researchers have begun to use of deep neural networks (DNN) to predict the dose distribution, engendering a new field of research [4][5][6][7][8][9][10][11][12]. Such dose prediction is useful for confirming the achievable dose distribution before or during the creation of treatment planning, and could reduce the iterative optimization process for IMRT, because the treatment planner can know which areas should receive increased or reduced doses based on the results of the prediction.…”

“…Although the DNN-based prediction models achieve good agreement between the predicted and original dose distributions, patient contours are necessary for the prediction in all the frameworks [4][5][6][7][8][9][10][11][12]. Structure contouring can be highly time-consuming: for example, an average time of approximately 4 hours is needed to contour for prostate treatment planning with Eclipse (Varian Medical Systems, Palo Alto, CA), and contouring for patients with head and neck cancer can take much longer [2].…”

Using deep learning to predict beam‐tunable Pareto optimal dose distribution for intensity‐modulated radiation therapy

“…Previous studies found that deep learning models that use anatomical structures and static beam orientations to predict Pareto optimal dose distributions could generate multiple optimal plans with differing trade-offs in real time. 44,55 However, Pareto optimal dose predictions for IMRT prostate plans with variable beam numbers and orientations have not yet been studied. In this paper, we present an approach that uses deep learning networks to predict Pareto optimal dose distributions for prostate IMRT plans that involve anatomical structures and varying beam numbers and orientations.…”

“…This contrasts with Pareto optimal dose prediction models, which can generate multiple plans that have differing trade‐offs between the different critical structures. Previous studies found that deep learning models that use anatomical structures and static beam orientations to predict Pareto optimal dose distributions could generate multiple optimal plans with differing trade‐offs in real time 44,55 . However, Pareto optimal dose predictions for IMRT prostate plans with variable beam numbers and orientations have not yet been studied.…”