Multi-Views Fusion CNN for Left Ventricular Volumes Estimation on Cardiac MR Images

Luo, Gongning; Dong, Siming; Wang, Kuanquan; Zuo, Wangmeng; Cao, Shaodong; Zhang, Henggui

doi:10.1109/tbme.2017.2762762

Cited by 70 publications

(43 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…So are most DCNN-based full heart segmentation methods [8] [9]. To process volumetric data, some models takes three perpendicular 2D slices as input and fuse the multi-view information abstraction for 3D segmentation [10,11]. 3D U-Net-like DCNNs, where the 2D operations were replaced by their 3D counterparts [12,13], were adopted in different applications.…”

Section: Introductionmentioning

confidence: 99%

A Two-Stage U-Net Model for 3D Multi-class Segmentation on Full-Resolution Cardiac Data

Wang

MacGillivray

Macnaught

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Deep convolutional neural networks (CNNs) have been intensively used for multi-class segmentation of data from different modalities and achieved state-of-the-art performances. However, a common problem when dealing with large, high resolution 3D data is that the volumes input into the deep CNNs has to be either cropped or downsampled due to limited memory capacity of computing devices. These operations lead to loss of resolution and increment of class imbalance in the input data batches, which can downgrade the performances of segmentation algorithms. Inspired by the architecture of image super-resolution CNN (SRCNN) and self-normalization network (SNN), we developed a two-stage modified Unet framework that simultaneously learns to detect a ROI within the full volume and to classify voxels without losing the original resolution. Experiments on a variety of multi-modal volumes demonstrated that, when trained with a simply weighted dice coefficients and our customized learning procedure, this framework shows better segmentation performances than state-of-the-art Deep CNNs with advanced similarity metrics.

show abstract

Section: Introductionmentioning

confidence: 99%

A Two-Stage U-Net Model for 3D Multi-class Segmentation on Full-Resolution Cardiac Data

Wang

MacGillivray

Macnaught

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…Due to accurate location can remove the most unrelative background, location is always an efficient process for accelerating small objects segmentation. Luo et al [3] utilized a combination of intersecting views and atlasbased method to locate the left ventricle. Ammar et al [4] located the LA by an empirical threshold method.…”

Section: Introductionmentioning

confidence: 99%

A Framework of Left Atrium Segmentation on CT Images with Combined Detection Network and Level Set Model

Liu¹,

Wang²,

Luo³

et al. 2019

2019 Computing in Cardiology Conference (CinC)

Self Cite

View full text Add to dashboard Cite

In this paper, we proposed a framework for left atrium (LA) segmentation on CT with combined detection network and level set model. The proposed framework consists of two steps. Firstly, we trained a Faster RCNN to generate location boxes for LA. The obtained location box can remove unrelated regions to reduce the interference of background and similarity tissues. Secondly, we utilized a self-adapted threshold on the location box to get the initialization for the level set model, which is nearer the LA and more robust than the random and fixed initialization. Then we proposed a 3D level set model with a new edge indicator based on DRLSE for the final LA segmentation. This edge indicator incorporated both numerical and direction information of the data gradient. Hence, the proposed level set model can guide the contour to the correct boundary when there are many boundaries surrounded the object. The framework was trained and evaluated on MICCAI 2013 LA segmentation challenge. The proposed segmentation method achieved the Dice score of 86.46%. Comparing to the original DRLSE, it achieved a 2.72% improvement on the Dice score.

show abstract

“…The statistical methods and machine learning methods all need a great deal of manpower and material resources, which is difficult for medical researches. Different from these traditional methods, deep learning methods, which can extract features automatically, have been successfully used for cardiac research [9]. [10] used a dynamic convolutional neural networks to segment fetal LV.…”

Section: Introductionmentioning

confidence: 99%

Weakly Supervised Deformation Network for 3D Echocardiography Segmentation on Left Ventricle

Dong¹,

Luo²,

Wulan³

et al. 2019

2019 Computing in Cardiology Conference (CinC)

Self Cite

View full text Add to dashboard Cite

The automated 3D echocardiography segmentation on left ventricle (LV) is very important for clinical evaluation of LV function. However, the segmentation is difficult due to the 3D echocardiography's challenges, such as the low signal-to-noise ratio, indistinguishable boundaries between LV and other heart substructures, and limited annotation data. This paper aims to propose a novel method to achieve accurate 3D echocardiography segmentation on LV, based on a weakly supervised deformable network. The deformation network was optimized by generative adversarial constraint and volume similarity constraint. The proposed framework was trained and validated on 3D echocardiography datasets which including 70 patients (35 train patients and 35 test patients). The results demonstrated the proposed method is relatively accurate and has potential for further research and application.

show abstract

Multi-Views Fusion CNN for Left Ventricular Volumes Estimation on Cardiac MR Images

Abstract: The proposed method not only has application potential for cardiac diseases screening for large-scale CMR data, but also can be extended to other medical image research fields.

Cited by 70 publications

References 43 publications

A Two-Stage U-Net Model for 3D Multi-class Segmentation on Full-Resolution Cardiac Data

A Two-Stage U-Net Model for 3D Multi-class Segmentation on Full-Resolution Cardiac Data

A Framework of Left Atrium Segmentation on CT Images with Combined Detection Network and Level Set Model

Weakly Supervised Deformation Network for 3D Echocardiography Segmentation on Left Ventricle

Contact Info

Product

Resources

About