An artificial intelligence-driven agent for real-time head-and-neck IMRT plan generation using conditional generative adversarial network (cGAN).

To develop an artificial intelligence (AI) agent for fully automated rapid head-and-neck intensity-modulated radiation therapy (IMRT) plan generation without time-consuming dose-volume-based inverse planning. This AI agent was trained via implementing a conditional generative adversarial network (cGAN) architecture. The generator, PyraNet, is a novel deep learning network that implements 28 classic ResNet blocks in pyramid-like concatenations. The discriminator is a customized four-layer DenseNet. The AI agent first generates multiple customized two-dimensional projections at nine template beam angles from a patient's three-dimensional computed tomography (CT) volume and structures. These projections are then stacked as four-dimensional inputs of PyraNet, from which nine radiation fluence maps of the corresponding template beam angles are generated simultaneously. Finally, the predicted fluence maps are automatically postprocessed by Gaussian deconvolution operations and imported into a commercial treatment planning system (TPS) for plan integrity check and visualization. The AI agent was built and tested upon 231 oropharyngeal IMRT plans from a TPS plan library. 200/16/15 plans were assigned for training/validation/testing, respectively. Only the primary plans in the sequential boost regime were studied. All plans were normalized to 44 Gy prescription (2 Gy/fx). A customized Harr wavelet loss was adopted for fluence map comparison during the training of the PyraNet. For test cases, isodose distributions in AI plans and TPS plans were qualitatively evaluated for overall dose distributions. Key dosimetric metrics were compared by Wilcoxon signed-rank tests with a significance level of 0.05. All 15 AI plans were successfully generated. Isodose gradients outside of PTV in AI plans were comparable to those of the TPS plans. After PTV coverage normalization, Dmean of left parotid (DAI  = 23.1 ± 2.4 Gy; DTPS  = 23.1 ± 2.0 Gy), right parotid (DAI  = 23.8 ± 3.0 Gy; DTPS  = 23.9 ± 2.3 Gy), and oral cavity (DAI  = 24.7 ± 6.0 Gy; DTPS  = 23.9 ± 4.3 Gy) in the AI plans and the TPS plans were comparable without statistical significance. AI plans achieved comparable results for maximum dose at 0.01cc of brainstem (DAI  = 15.0 ± 2.1 Gy; DTPS  = 15.5 ± 2.7 Gy) and cord + 5mm (DAI  = 27.5 ± 2.3 Gy; DTPS  = 25.8 ± 1.9 Gy) without clinically relevant differences, but body Dmax results (DAI  = 121.1 ± 3.9 Gy; DTPS  = 109.0 ± 0.9 Gy) were higher than the TPS plan results. The AI agent needed ~3 s for predicting fluence maps of an IMRT plan. With rapid and fully automated execution, the developed AI agent can generate complex head-and-neck IMRT plans with acceptable dosimetry quality. This approach holds great potential for clinical applications in preplanning decision-making and real-time planning.

Li X ，Wang C ，Sheng Y ，Zhang J ，Wang W ，Yin FF ，Wu Q ，Wu QJ ，Ge Y ... -  《-》

Automatic IMRT planning via static field fluence prediction (AIP-SFFP): a deep learning algorithm for real-time prostate treatment planning.

Li X ，Zhang J ，Sheng Y ，Chang Y ，Yin FF ，Ge Y ，Wu QJ ，Wang C ... -  《-》

Attention-aware 3D U-Net convolutional neural network for knowledge-based planning 3D dose distribution prediction of head-and-neck cancer.

Deep learning-based knowledge-based planning (KBP) methods have been introduced for radiotherapy dose distribution prediction to reduce the planning time and maintain consistent high-quality plans. This paper presents a novel KBP model using an attention-gating mechanism and a three-dimensional (3D) U-Net for intensity-modulated radiation therapy (IMRT) 3D dose distribution prediction in head-and-neck cancer. A total of 340 head-and-neck cancer plans, representing the OpenKBP-2020 AAPM Grand Challenge data set, were used in this study. All patients were treated with the IMRT technique and a dose prescription of 70 Gy. The data set was randomly divided into 64%/16%/20% as training/validation/testing cohorts. An attention-gated 3D U-Net architecture model was developed to predict full 3D dose distribution. The developed model was trained using the mean-squared error loss function, Adam optimization algorithm, a learning rate of 0.001, 120 epochs, and batch size of 4. In addition, a baseline U-Net model was also similarly trained for comparison. The model performance was evaluated on the testing data set by comparing the generated dose distributions against the ground-truth dose distributions using dose statistics and clinical dosimetric indices. Its performance was also compared to the baseline model and the reported results of other deep learning-based dose prediction models. The proposed attention-gated 3D U-Net model showed high capability in accurately predicting 3D dose distributions that closely replicated the ground-truth dose distributions of 68 plans in the test set. The average value of the mean absolute dose error was 2.972 ± 1.220 Gy (vs. 2.920 ± 1.476 Gy for a baseline U-Net) in the brainstem, 4.243 ± 1.791 Gy (vs. 4.530 ± 2.295 Gy for a baseline U-Net) in the left parotid, 4.622 ± 1.975 Gy (vs. 4.223 ± 1.816 Gy for a baseline U-Net) in the right parotid, 3.346 ± 1.198 Gy (vs. 2.958 ± 0.888 Gy for a baseline U-Net) in the spinal cord, 6.582 ± 3.748 Gy (vs. 5.114 ± 2.098 Gy for a baseline U-Net) in the esophagus, 4.756 ± 1.560 Gy (vs. 4.992 ± 2.030 Gy for a baseline U-Net) in the mandible, 4.501 ± 1.784 Gy (vs. 4.925 ± 2.347 Gy for a baseline U-Net) in the larynx, 2.494 ± 0.953 Gy (vs. 2.648 ± 1.247 Gy for a baseline U-Net) in the PTV_70, and 2.432 ± 2.272 Gy (vs. 2.811 ± 2.896 Gy for a baseline U-Net) in the body contour. The average difference in predicting the D99 value for the targets (PTV_70, PTV_63, and PTV_56) was 2.50 ± 1.77 Gy. For the organs at risk, the average difference in predicting the D m a x ${D_{max}}$ (brainstem, spinal cord, and mandible) and D m e a n ${D_{mean}}$ (left parotid, right parotid, esophagus, and larynx) values was 1.43 ± 1.01 and 2.44 ± 1.73 Gy, respectively. The average value of the homogeneity index was 7.99 ± 1.45 for the predicted plans versus 5.74 ± 2.95 for the ground-truth plans, whereas the average value of the conformity index was 0.63 ± 0.17 for the predicted plans versus 0.89 ± 0.19 for the ground-truth plans. The proposed model needs less than 5 s to predict a full 3D dose distribution of 64 × 64 × 64 voxels for a new patient that is sufficient for real-time applications. The attention-gated 3D U-Net model demonstrated a capability in predicting accurate 3D dose distributions for head-and-neck IMRT plans with consistent quality. The prediction performance of the proposed model was overall superior to a baseline standard U-Net model, and it was also competitive to the performance of the best state-of-the-art dose prediction method reported in the literature. The proposed model could be used to obtain dose distributions for decision-making before planning, quality assurance of planning, and guiding-automated planning for improved plan consistency, quality, and planning efficiency.

Osman AFI ，Tamam NM 《Journal of Applied Clinical Medical Physics》

Insights of an AI agent via analysis of prediction errors: a case study of fluence map prediction for radiation therapy planning.

Purpose.We have previously reported an artificial intelligence (AI) agent that automatically generates intensity-modulated radiation therapy (IMRT) plans via fluence map prediction, by-passing inverse planning. This AI agent achieved clinically comparable quality for prostate cases, but its performance on head-and-neck patients leaves room for improvement. This study aims to collect insights of the deep-learning-based (DL-based) fluence map prediction model by systematically analyzing its prediction errors.Methods.From the modeling perspective, the DL model's output is the fluence maps of IMRT plans. However, from the clinical planning perspective, the plan quality evaluation should be based on the clinical dosimetric criteria such as dose-volume histograms. To account for the complex and non-intuitive relationships between fluence map prediction errors and the corresponding dose distribution changes, we propose a novel error analysis approach that systematically examines plan dosimetric changes that are induced by varying amounts of fluence prediction errors. We investigated four decomposition modes of model prediction errors. The two spatial domain decompositions are based on fluence intensity and fluence gradient. The two frequency domain decompositions are based on Fourier-space banded frequency rings and Fourier-space truncated low-frequency disks. The decomposed error was analyzed for its impact on the resulting plans' dosimetric metrics. The analysis was conducted on 15 test cases spared from the 200 training and 16 validation cases used to train the model.Results.Most planning target volume metrics were significantly correlated with most error decompositions. The Fourier space disk radii had the largest Spearman's coefficients. The low-frequency region within a disk of ∼20% Fourier space contained most of errors that impact overall plan quality.Conclusions.This study demonstrates the feasibility of using fluence map prediction error analysis to understand the AI agent's performance. Such insights will help fine-tune the DL models in architecture design and loss function selection.

Li X ，Wu QJ ，Wu Q ，Wang C ，Sheng Y ，Wang W ，Stephens H ，Yin FF ，Ge Y ... -  《-》

Dosimetric comparison of two treatment planning systems for spine SBRT.

Two commonly used treatment planning systems (TPS) are compared for the planning of spine stereotactic body radiotherapy (SBRT). The main purpose is to highlight relative advantages and disadvantages of each system and propose a methodologic approach for comparisons. Twenty clinical plans were inversely planned with step-and-shoot intensity-modulated radiotherapy (IMRT) each using 9 to 11 beams, referred to as IMRT_P. The prescription dose was 24 Gy in 2 fractions, and the plans were generated following our institutional protocol using the Pinnacle3 v9.2. Each case was replanned using a 2-arc volumetric modulated arc therapy (VMAT) approach, referred as VMAT_P. CT and structure sets were DICOM exported to Monaco v5.10 and planned in 2 different ways: IMRT (IMRT_M) and VMAT (VMAT_M) using the same prescription dose. Dose volume histograms (DVH) and other dose statistics of planning target volumes (PTV) and organ-at-risk (OAR) were analyzed and compared between plans. The gradient index (GI = ratio of 50% isodose volume to prescribed isodose volume) was used to measure dose fall-off outside of the PTV. Another metric - Gradient Index Inner (GIinner = the rate (in Gy/mm) - at which the dose changes from the level of the spinal cord/thecal sac toward the prescription dose) was developed and compared. All plans were considered clinically acceptable by institutional guidelines and achieved all of the OAR dose constraints. VMAT_M and IMRT_M showed comparable dose statistics for the PTV when compared to VMAT_P and IMRT_P, respectively. For IMRT plans, the median GIinner was 1.88 Gy/mm vs 1.52 Gy/mm for IMRT_M and IMRT_P respectively (p< 0.001). All other IMRT metrics were statistically similar except for the PTV maximum dose (Dmax), which was higher for IMRT_M than IMRT_P (median 30.7 Gy vs 29.0 Gy, p< 0.001). For VMAT plans, only PTV Dmin showed a statistical different between VMAT_M and VMAT_P of median 12.7 Gy vs 9.7 Gy (p< 0.001). In terms of beam sequencing parameters, the number of monitor units was statistically higher for VMAT_P compared to VMAT_M (median = 6764 vs 5376) whereas the number of segments for IMRT_M was statistically greater than IMRT_P (median = 155 vs 73). We were able to generate clinically acceptable plans for different types of spine SBRT using 2 different TPS. We used an evaluation strategy involving coverage, conformity, and dose gradient that can compared between TPS.

Lee YK ，Munawar I ，Mashouf S ，Sahgal A ，Ruschin M ... -  《-》