Search

BackgroundMyasthenia gravis (MG) is the most common paraneoplastic syndromes of thymoma and closely related to thymus abnormalities. Timely detecting of the risk of MG would benefit clinical management and treatment decision for patients with thymoma. Herein, we developed a 3D DenseNet deep learning (DL) model based on preoperative computed tomography (CT) as a non-invasive method to detect MG in thymoma patients.MethodsA large cohort of 230 thymoma patients in a hospital affiliated with a medical school were enrolled. 182 thymoma patients (81 with MG, 101 without MG) were used for training and model building. 48 cases from another hospital were used for external validation. A 3D-DenseNet-DL model and five radiomic models were performed to detect MG in thymoma patients. A comprehensive analysis by integrating machine learning and semantic CT image features, named 3D-DenseNet-DL-based multi-model, was also performed to establish a more effective prediction model.FindingsBy elaborately comparing the prediction efficacy, the 3D-DenseNet-DL effectively identified MG patients and was superior to other five radiomic models, with a mean area under ROC curve (AUC), accuracy, sensitivity, and specificity of 0.734, 0.724, 0.787, and 0.672, respectively. The effectiveness of the 3D-DenseNet-DL-based multi-model was further improved as evidenced by the following metrics: AUC 0.766, accuracy 0.790, sensitivity 0.739, and specificity 0.801. External verification results confirmed the feasibility of this DL-based multi-model with metrics: AUC 0.730, accuracy 0.732, sensitivity 0.700, and specificity 0.690, respectively.InterpretationOur 3D-DenseNet-DL model can effectively detect MG in patients with thymoma based on preoperative CT imaging. This model may serve as a supplement to the conventional diagnostic criteria for identifying thymoma associated MG.

Purpose: The use of convolution neural networks (CNN) to accurately predict dose distributions can accelerate intensity-modulated radiation therapy (IMRT) planning. The purpose of our study is to develop a novel deep learning architecture for precise voxel-level dose prediction on brain tumors. Methods: A dataset of 120 patients with brain tumors is built for the retrospective study. The dose distributions are predicted by a designed end-to-end model called TS-Net, in which the transformer encoder module is utilized to obtain abundant global features by learning long-range correlations of the input sequence. In addition, semantic field alignment (SFA) block is proposed in decoding path to ensure effective propagation of strong semantic information from deep to shallow. Five images from different channels are fed into the architecture, including a computed tomography (CT) image, a planning target volumes (PTV) image, an organs-at-risk (OARs) image, a beam configuration image, and a distance image, and the predicted dose distributions are taken as outputs. We use different evaluation metrics to evaluate the performance of the model and discuss the role of the auxiliary beam configuration information provided by non-modulated dose distributions. Results: The TS-Net prediction accuracies in terms of mean absolute error (MAE) are 2.98% for PTV, 7.19% for brainstem, 1.88% for left len, 2.48% for right len, 9.61% for left optic nerve, 9.10% for right optic nerve, 8.99% for optic chiasma, and 8.28% for pituitary. There is no statistically significant difference between the predicted results and clinical dose distributions for clinical indexes including homogeneity index (HI), D50, and D95 for PTV; V40, mean dose, and max dose for OARs; except for conformation index (CI) and D2 for PTV. The model has dice similarity coefficient (DSC) values of above 0.91 for most isodose volumes, clearly outperforming HD U-Net, and being slightly better than U-Net and DCNN. Conclusion: The proposed TS-Net with beam configuration input can achieve accurate voxel-level dose prediction for brain tumors, and is a usable tool for improving the efficiency and quality of radiotherapy. [ABSTRACT FROM AUTHOR]