RESUMO
Background: Pulmonary epithelioid hemangioendothelioma (PEH) is a rare vascular tumour, and its early diagnosis remains challenging. This study aims to comprehensively analyse the imaging features of PEH and develop a model for predicting PEH. Methods: Retrospective and pooled analyses of imaging findings were performed in PEH patients at our center (n=25) and in published cases (n=71), respectively. Relevant computed tomography (CT) images were extracted and used to build a deep learning model for PEH identification and differentiation from other diseases. Results: In this study, bilateral multiple nodules/masses (n=19) appeared to be more common with most nodules less than 2 cm. In addition to the common types and features, the pattern of mixed type (n=4) and isolated nodules (n=4), punctate calcifications (5/25) and lymph node enlargement were also observed (10/25). The presence of pleural effusion is associated with a poor prognosis in PEH. The deep learning model, with an area under the receiver operating characteristic curve (AUC) of 0.71 [95% confidence interval (CI): 0.69-0.72], has a differentiation accuracy of 100% and 74% for the training and test sets respectively. Conclusions: This study confirmed the heterogeneity of the imaging findings in PEH and showed several previously undescribed types and features. The current deep learning model based on CT has potential for clinical application and needs to be further explored in the future.
RESUMO
Background.Concern has been expressed regarding the risk of carcinogenesis from medical computed tomography (CT) radiation. Lowering radiation in CT without appropriate modifications often leads to severe noise-induced artifacts in the images. The utilization of deep learning (DL) techniques has achieved promising reconstruction performance in low-dose CT (LDCT) imaging. However, most DL-based algorithms require the pre-collection of a large set of image pairs (low-dose/standard-dose) and the training of networks in an end-to-end supervised manner. Meanwhile, securing such a large volume of paired, well-registered training data in clinical practice is challenging. Moreover, these algorithms often overlook the potential to utilize the abundant information in a large collection of LDCT-only images/sinograms.Methods.In this paper, we introduce a semi-supervised iterative adaptive network (SIA-Net) for LDCT imaging, utilizing both labeled and unlabeled sinograms in a cohesive network framework, integrating supervised and unsupervised learning processes. Specifically, the supervised process captures critical features (i.e. noise distribution and tissue characteristics) latent in the paired sinograms, while the unsupervised process effectively learns these features in the unlabeled low-dose sinograms, employing a conventional weighted least-squares model with a regularization term. Furthermore, the SIA-Net method is designed to adaptively transfer the learned feature distribution from the supervised to the unsupervised process, thereby obtaining a high-fidelity sinogram through iterative adaptive learning. Finally, high-quality CT images can be reconstructed from the refined sinogram using the filtered back-projection algorithm.Results.Experimental results on two clinical datasets indicate that the proposed SIA-Net method achieves competitive performance in terms of noise reduction and structure preservation in LDCT imaging, when compared to traditional supervised learning methods.