2015
DOI: 10.1007/978-3-319-24553-9_82
|View full text |Cite
|
Sign up to set email alerts
|

Direct and Simultaneous Four-Chamber Volume Estimation by Multi-Output Regression

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
32
0

Year Published

2016
2016
2019
2019

Publication Types

Select...
6
2

Relationship

2
6

Authors

Journals

citations
Cited by 36 publications
(32 citation statements)
references
References 10 publications
0
32
0
Order By: Relevance
“…Two-phase framework employed in existing direct methods [6]- [11] is inadequate to achieve accurate estimation for Right: the cardiac indices to be estimated in this work, which include area of LV cavity (A1), area of myocardium (A2), and the 6 regional wall thickness (WT1∼WT6). multitype cardiac indices, for the reason that image representation and indices regression are separately handled, and no feedback connection exists between them during optimization.…”
Section: B Existing Two-phase Direct Methodsmentioning
confidence: 99%
See 1 more Smart Citation
“…Two-phase framework employed in existing direct methods [6]- [11] is inadequate to achieve accurate estimation for Right: the cardiac indices to be estimated in this work, which include area of LV cavity (A1), area of myocardium (A2), and the 6 regional wall thickness (WT1∼WT6). multitype cardiac indices, for the reason that image representation and indices regression are separately handled, and no feedback connection exists between them during optimization.…”
Section: B Existing Two-phase Direct Methodsmentioning
confidence: 99%
“…These handcrafted features were further replaced with a more effective image representation that learned from a multiscale convolutional deep belief network (MCDBN) [10], leading to improved correlation between the estimated volumes and their ground truth. Supervised descriptor learning (SDL) was proposed in the work of four chamber volumes estimation [11], which still employed a separate adaptive K-clustering random forest (AKRF) regression [16]. The compatibility between the descriptor and the regression model still can not be enhanced in the two-phase framework.…”
Section: B Existing Two-phase Direct Methodsmentioning
confidence: 99%
“…In these methods, hand-crafted features extracted from cardiac images are fed into regression models such as random forest (RF), adaptive K-clustering RF (AKRF), Bayesian model, and neural networks, to predict cardiac volumes. The employed features include Bhattacharyya coefficient between image distributions [1,2], appearance features [18], multiple low level image features [21], as well as features from multiscale convolutional deep belief network (MCDNB) [22] and supervised descriptor learning (SDL) [20]. Although these methods obtained effective performance, two limitations still exist: 1) they followed two separated phases, i.e., feature extraction+ volumes regression, and no feedback exists between them to make them compatible with each other; 2) neither the temporal dependencies nor the spatial dependencies are taken into account, while the dependencies are important for dynamic modeling of cardiac sequence.…”
Section: Related Workmentioning
confidence: 99%
“…To demonstrate the advantages of our proposed method over segmentation based [3] and two-phase direct methods [21,22,20], we apply these methods to our database for cardiac RWT estimation. For the direct methods, the same five-fold crossvalidation protocol is used.…”
Section: Performance Comparison: Resrnn Vs State-of-the-artmentioning
confidence: 99%
“…5,6,[16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31] But LV segmentation is still an open problem and is challenging due to the poor contrast between tissues around the epicardium and image artifacts, such as intensity inhomogeneities in cine cardiac magnetic resonance (CMR) images. 6,12,19,[31][32][33][34][35][36][37] Traditional segmentation techniques, such as thresholding and k-means clustering, have been used to segment the LV from cardiac MR images. [38][39][40][41] Active shape modeling and active appearance modeling have also become prominent tools in LV segmentation.…”
Section: Introductionmentioning
confidence: 99%