The segmentation of unlabeled medical images is troublesome due to the high cost of annotation, and unsupervised domain adaptation is one solution to this. In this paper, an improved unsupervised domain adaptation met...The segmentation of unlabeled medical images is troublesome due to the high cost of annotation, and unsupervised domain adaptation is one solution to this. In this paper, an improved unsupervised domain adaptation method was proposed. The proposed method considered both global alignment and category-wise alignment. First, we aligned the appearance of two domains by image transformation. Second, we aligned the output maps of two domains in a global way. Then, we decomposed the semantic prediction map by category, aligning the prediction maps in a category-wise manner. Finally, we evaluated the proposed method on the 2017 Multi-Modality Whole Heart Segmentation Challenge dataset, and obtained 82.1 on the dice similarity coefficient and 4.6 on the average symmetric surface distance, demonstrating the effectiveness of the combination of global alignment and category-wise alignment.展开更多
Analyzing cardiac pathology using image-derived features is a complex undertaking requiring a substantial amount of labeled data.Semi-supervised learning requires a small amount of labeled data and a large amount of u...Analyzing cardiac pathology using image-derived features is a complex undertaking requiring a substantial amount of labeled data.Semi-supervised learning requires a small amount of labeled data and a large amount of unlabeled data.However,most semi-supervised approaches are not robust for ventricular segmentation to extract shape-based information;hence,they have limited segmentation accuracy compared with supervised learning.Therefore,we proposed a dual-path copy-paste segmentation network in a mean Teacher architecture to learn knowledge from labeled images and transfer it to unlabeled images in a dual-path learning manner.This effectively reduces the empirical data distribution gap and learns the semantic information from labeled data in both the inward and outward directions for shape-based feature extraction.We also extracted the motion parameters from two input image frames(2D+)of the same slice from a cine magnetic resonance image(MRI).After that,we fused the segmentation mask of the myocardium wall and motion parameters to generate the dynamic characteristics of the time-series for cardiac pathology classification.In our evaluation,we compared the segmentation outcomes of our method with a state-of-the-art semi-supervised approach using the automatic cardiac diagnosis challenge(ACDC)dataset.We assessed two different ratios of labeled data,at 5%and 10%,respectively.Our method delivered impressive results,achieving an average dice score of 87.85%and 88.66%under these conditions.Moreover,our model shows a promising classification accuracy of 97%for the training set and 96%for the testing set.The proposed method was trained end-to-end,demonstrating a general framework for automatic cardiac pathology classification from cine MRI in a clinical setting.展开更多
An infant male presented with the rare anatomy consisting of situs solitus,concordant atrioventricular connections to L-looped ventricles,double outlet right ventricle(DORV),and hypoplastic aortic arch.6 months after ...An infant male presented with the rare anatomy consisting of situs solitus,concordant atrioventricular connections to L-looped ventricles,double outlet right ventricle(DORV),and hypoplastic aortic arch.6 months after neonatal aortic arch repair,the morphologic right ventricle function deteriorated,and surgical evaluation was undertaken to determine if either biventricular repair with a systemic morphologic left ventricle or right ventricular exclusion was possible.After initial echocardiography,magnetic resonance imaging(MRI)was used to create detailed axial and 4-dimensional(4D)images and 3-dimensional(3D)printed models.The detailed anatomy of this rare,complex case and its use in pre-surgical planning is presented.展开更多
文摘The segmentation of unlabeled medical images is troublesome due to the high cost of annotation, and unsupervised domain adaptation is one solution to this. In this paper, an improved unsupervised domain adaptation method was proposed. The proposed method considered both global alignment and category-wise alignment. First, we aligned the appearance of two domains by image transformation. Second, we aligned the output maps of two domains in a global way. Then, we decomposed the semantic prediction map by category, aligning the prediction maps in a category-wise manner. Finally, we evaluated the proposed method on the 2017 Multi-Modality Whole Heart Segmentation Challenge dataset, and obtained 82.1 on the dice similarity coefficient and 4.6 on the average symmetric surface distance, demonstrating the effectiveness of the combination of global alignment and category-wise alignment.
基金supported by the Chinese Academy of Sciences Youth Innovation Promotion Association Excellent Member Program,China(No.Y201968).
文摘Analyzing cardiac pathology using image-derived features is a complex undertaking requiring a substantial amount of labeled data.Semi-supervised learning requires a small amount of labeled data and a large amount of unlabeled data.However,most semi-supervised approaches are not robust for ventricular segmentation to extract shape-based information;hence,they have limited segmentation accuracy compared with supervised learning.Therefore,we proposed a dual-path copy-paste segmentation network in a mean Teacher architecture to learn knowledge from labeled images and transfer it to unlabeled images in a dual-path learning manner.This effectively reduces the empirical data distribution gap and learns the semantic information from labeled data in both the inward and outward directions for shape-based feature extraction.We also extracted the motion parameters from two input image frames(2D+)of the same slice from a cine magnetic resonance image(MRI).After that,we fused the segmentation mask of the myocardium wall and motion parameters to generate the dynamic characteristics of the time-series for cardiac pathology classification.In our evaluation,we compared the segmentation outcomes of our method with a state-of-the-art semi-supervised approach using the automatic cardiac diagnosis challenge(ACDC)dataset.We assessed two different ratios of labeled data,at 5%and 10%,respectively.Our method delivered impressive results,achieving an average dice score of 87.85%and 88.66%under these conditions.Moreover,our model shows a promising classification accuracy of 97%for the training set and 96%for the testing set.The proposed method was trained end-to-end,demonstrating a general framework for automatic cardiac pathology classification from cine MRI in a clinical setting.
文摘An infant male presented with the rare anatomy consisting of situs solitus,concordant atrioventricular connections to L-looped ventricles,double outlet right ventricle(DORV),and hypoplastic aortic arch.6 months after neonatal aortic arch repair,the morphologic right ventricle function deteriorated,and surgical evaluation was undertaken to determine if either biventricular repair with a systemic morphologic left ventricle or right ventricular exclusion was possible.After initial echocardiography,magnetic resonance imaging(MRI)was used to create detailed axial and 4-dimensional(4D)images and 3-dimensional(3D)printed models.The detailed anatomy of this rare,complex case and its use in pre-surgical planning is presented.
