Tissues margin-based analytical anisotropic algorithm boosting method via deep learning attention mechanism with cervical cancer.

PURPOSE: Speed and accuracy are two critical factors in dose calculation for radiotherapy. Analytical Anisotropic Algorithm (AAA) is a rapid dose calculation algorithm but has dose errors in tissue margin area. Acuros XB (AXB) has high accuracy but takes long time to calculate. To improve the dose accuracy on the tissue margin area for AAA, we proposed a novel deep learning-based dose accuracy improvement method using Margin-Net combined with Margin-Loss. METHODS: A novel model 'Margin-Net' was designed with a Margin Attention Mechanism to generate special margin-related features. Margin-Loss was introduced to consider the dose errors and dose gradients in tissues margin area. Ninety-five VMAT cervical cancer cases with paired AAA and AXB dose were enrolled in our study: 76 cases for training and 19 cases for testing. Tissues Margin Masks were generated from RT contours with 6 mm extension. Tissues Margin Mask, AAA dose and CTs were input data; AXB dose was used as reference dose for model training and evaluation. Comparison experiments were performed to evaluated effectiveness of Margin-Net and Margin-Loss. RESULTS: Compared to AXB dose, the 3D gamma passing rate (1%/1 mm, 10% threshold) for 19 test cases 95.75 ± 1.05% using Margin-Net with Margin-Loss, which was significantly higher than the original AAA dose (73.64 ± 3.46%). The passing rate reduced to 94.07 ± 1.16% without Margin-Loss and 87.3 ± 1.18% if Margin-Net key structure 'MAM' was also removed. CONCLUSION: The proposed novel tissues margin-based dose conversion method can significantly improve the dose accuracy of Analytical Anisotropic Algorithm to be comparable to AXB algorithm. It can potentially improve the efficiency of treatment planning process with low demanding of computation resources.

Accuracy Improvement Method Based on Characteristic Database Classification for IMRT Dose Prediction in Cervical Cancer: Scientifically Training Data Selection.

Purpose: Consistent training and testing datasets can lead to good performance for deep learning (DL) models. However, a large high-quality training dataset for unusual clinical scenarios is usually not easy to collect. The work aims to find optimal training data collection strategies for DL-based dose prediction models. Materials and Methods: A total of 325 clinically approved cervical IMRT plans were utilized. We designed comparison experiments to investigate the impact of (1) beam angles, (2) the number of beams, and (3) patient position for DL dose prediction models. In addition, a novel geometry-based beam mask generation method was proposed to provide beam setting information in the model training process. What is more, we proposed a new training strategy named "full-database pre-trained strategy". Results: The model trained with a homogeneous dataset with the same beam settings achieved the best performance [mean prediction errors of planning target volume (PTV), bladder, and rectum: 0.29 ± 0.15%, 3.1 ± 2.55%, and 3.15 ± 1.69%] compared with that trained with large mixed beam setting plans (mean errors of PTV, bladder, and rectum: 0.8 ± 0.14%, 5.03 ± 2.2%, and 4.45 ± 1.4%). A homogeneous dataset is more accessible to train an accurate dose prediction model (mean errors of PTV, bladder and rectum: 2.2 ± 0.15%, 5 ± 2.1%, and 3.23 ± 1.53%) than a non-homogeneous one (mean errors of PTV, bladder and rectum: 2.55 ± 0.12%, 6.33 ± 2.46%, and 4.76 ± 2.91%) without other processing approaches. The added beam mask can constantly improve the model performance, especially for datasets with different beam settings (mean errors of PTV, bladder, and rectum improved from 0.8 ± 0.14%, 5.03 ± 2.2%, and 4.45 ± 1.4% to 0.29 ± 0.15%, 3.1 ± 2.55%, and 3.15 ± 1.69%). Conclusions: A consistent dataset is recommended to form a patient-specific IMRT dose prediction model. When a consistent dataset is not accessible to collect, a large dataset with different beam angles and a training model with beam information can also get a relatively good model. The full-database pre-trained strategies can rapidly form an accuracy model from a pre-trained model. The proposed beam mask can effectively improve the model performance. Our study may be helpful for further dose prediction studies in terms of training strategies or database establishment.

Evaluation Exploration of Atlas-Based and Deep Learning-Based Automatic Contouring for Nasopharyngeal Carcinoma.

Purpose: The purpose of this study was to evaluate and explore the difference between an atlas-based and deep learning (DL)-based auto-segmentation scheme for organs at risk (OARs) of nasopharyngeal carcinoma cases to provide valuable help for clinical practice. Methods: 120 nasopharyngeal carcinoma cases were established in the MIM Maestro (atlas) database and trained by a DL-based model (AccuContour®), and another 20 nasopharyngeal carcinoma cases were randomly selected outside the atlas database. The experienced physicians contoured 14 OARs from 20 patients based on the published consensus guidelines, and these were defined as the reference volumes (Vref). Meanwhile, these OARs were auto-contoured using an atlas-based model, a pre-built DL-based model, and an on-site trained DL-based model. These volumes were named Vatlas, VDL-pre-built, and VDL-trained, respectively. The similarities between Vatlas, VDL-pre-built, VDL-trained, and Vref were assessed using the Dice similarity coefficient (DSC), Jaccard coefficient (JAC), maximum Hausdorff distance (HDmax), and deviation of centroid (DC) methods. A one-way ANOVA test was carried out to show the differences (between each two of them). Results: The results of the three methods were almost similar for the brainstem and eyes. For inner ears and temporomandibular joints, the results of the pre-built DL-based model are the worst, as well as the results of atlas-based auto-segmentation for the lens. For the segmentation of optic nerves, the trained DL-based model shows the best performance (p < 0.05). For the contouring of the oral cavity, the DSC value of VDL-pre-built is the smallest, and VDL-trained is the most significant (p < 0.05). For the parotid glands, the DSC of Vatlas is the minimum (about 0.80 or so), and VDL-pre-built and VDL-trained are slightly larger (about 0.82 or so). In addition to the oral cavity, parotid glands, and the brainstem, the maximum Hausdorff distances of the other organs are below 0.5 cm using the trained DL-based segmentation model. The trained DL-based segmentation method behaves well in the contouring of all the organs that the maximum average deviation of the centroid is no more than 0.3 cm. Conclusion: The trained DL-based segmentation performs significantly better than atlas-based segmentation for nasopharyngeal carcinoma, especially for the OARs with small volumes. Although some delineation results still need further modification, auto-segmentation methods improve the work efficiency and provide a level of help for clinical work.

Dose Prediction Using a Three-Dimensional Convolutional Neural Network for Nasopharyngeal Carcinoma With Tomotherapy.

PURPOSE: This study focused on predicting 3D dose distribution at high precision and generated the prediction methods for nasopharyngeal carcinoma patients (NPC) treated with Tomotherapy based on the patient-specific gap between organs at risk (OARs) and planning target volumes (PTVs). METHODS: A convolutional neural network (CNN) is trained using the CT and contour masks as the input and dose distributions as output. The CNN is based on the "3D Dense-U-Net", which combines the U-Net and the Dense-Net. To evaluate the model, we retrospectively used 124 NPC patients treated with Tomotherapy, in which 96 and 28 patients were randomly split and used for model training and test, respectively. We performed comparison studies using different training matrix shapes and dimensions for the CNN models, i.e., 128 ×128 ×48 (for Model I), 128 ×128 ×16 (for Model II), and 2D Dense U-Net (for Model III). The performance of these models was quantitatively evaluated using clinically relevant metrics and statistical analysis. RESULTS: We found a more considerable height of the training patch size yields a better model outcome. The study calculated the corresponding errors by comparing the predicted dose with the ground truth. The mean deviations from the mean and maximum doses of PTVs and OARs were 2.42 and 2.93%. Error for the maximum dose of right optic nerves in Model I was 4.87 ± 6.88%, compared with 7.9 ± 6.8% in Model II (p=0.08) and 13.85 ± 10.97% in Model III (p<0.01); the Model I performed the best. The gamma passing rates of PTV60 for 3%/3 mm criteria was 83.6 ± 5.2% in Model I, compared with 75.9 ± 5.5% in Model II (p<0.001) and 77.2 ± 7.3% in Model III (p<0.01); the Model I also gave the best outcome. The prediction error of D95 for PTV60 was 0.64 ± 0.68% in Model I, compared with 2.04 ± 1.38% in Model II (p<0.01) and 1.05 ± 0.96% in Model III (p=0.01); the Model I was also the best one. CONCLUSIONS: It is significant to train the dose prediction model by exploiting deep-learning techniques with various clinical logic concepts. Increasing the height (Y direction) of training patch size can improve the dose prediction accuracy of tiny OARs and the whole body. Our dose prediction network model provides a clinically acceptable result and a training strategy for a dose prediction model. It should be helpful to build automatic Tomotherapy planning.