Knowledge-Based Multi-sequence MR Segmentation via Deep Learning with a Hybrid U-Net++ Model

Ren, Jinchang; Sun, He; Huang, Yumin; Gao, Hao

doi:10.1007/978-3-030-39074-7_30

Cited by 6 publications

(3 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The number of intermediary skip connection units depends on the layer number and decreases linearly when traversing the contracting path. Applications in U-net++ include segmentation of cell nuclei [73], cancer tissue [73], cardiac structures and vessels [74], [75], and pelvic organs [76].…”

Section: H U-net++mentioning

confidence: 99%

See 1 more Smart Citation

U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications

et al. 2021

View full text Add to dashboard Cite

U-net is an image segmentation technique developed primarily for image segmentation tasks. These traits provide U-net with a high utility within the medical imaging community and have resulted in extensive adoption of U-net as the primary tool for segmentation tasks in medical imaging. The success of U-net is evident in its widespread use in nearly all major image modalities, from CT scans and MRI to Xrays and microscopy. Furthermore, while U-net is largely a segmentation tool, there have been instances of the use of U-net in other applications. Given that U-net's potential is still increasing, this narrative literature review examines the numerous developments and breakthroughs in the U-net architecture and provides observations on recent trends. We also discuss the many innovations that have advanced in deep learning and discuss how these tools facilitate U-net. In addition, we review the different image modalities and application areas that have been enhanced by U-net.

show abstract

Section: H U-net++mentioning

confidence: 99%

“…U-net implementation has also been applied on cardiovascular MR images [4]- [6], [8]- [11], [47], [58], [62], [74], [83], [86], [90], [137]- [146] to segment structures of the heart. Cancer is a leading cause of deaths worldwide and MR is one of the strongest methods for proper prognosis of various types of cancers.…”

Section: Magnetic Resonance Imaging (Mri)mentioning

confidence: 99%

U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications

et al. 2021

View full text Add to dashboard Cite

show abstract

“…In the last few years, the encoder-decoder based deep learning architectures, such as the U-Net [3] and FCN [4], have achieved a state-of-the-art performance in image segmentation. Thanks to its strong efficiency and a simple structure, it becomes a trend to apply the U-Net based models for a variety of applications in medical image segmentation, such as the segmentation of retinal vessels [5], brain tumors [6,7], cells [8], lesions [10], the pulmonary lobes [11], and the cardiac MRI [1,2,9]. With the interactions between the encoder path and the decoder path, the U-Net model can efficiently combine both the low-level and high-level features with several down-sampling, up-sampling and a specific skip connection for discovering more fine-grained details.…”

Section: Introductionmentioning

confidence: 99%

Effective extraction of ventricles and myocardium objects from cardiac magnetic resonance images with a multi-task learning U-Net

Ren

Sun

Zhao

et al. 2022

Pattern Recognition Letters

View full text Add to dashboard Cite

0.7 Å Resolution Electron Tomography Enabled by Deep‐Learning‐Aided Information Recovery

Wang

Ding

Liu

et al. 2020

Advanced Intelligent Systems

View full text Add to dashboard Cite

As the architectural complexity of integrated circuits and functional nanomaterials advances into the atomic-scale regimes, high-resolution transmission electron microscopy (TEM) has become an essential tool in validating the structure of nanomaterials and devices. In contrast to conventional TEM imaging, which only provides projected 2D information, electron tomography is a powerful tool that can probe the 3D internal structure and chemistry of materials at the nanoscale and the atomic scale. [1] Due to its high resolution, it has been widely used in biological, [2,3] physical, [4,5] and materials science [6-10] applications. Recently, with the development of discrete tomography [11-13] and atomic-scale electron tomography, [14-18] deciphering the structure of materials atom by atom has become possible. Electron tomography renders 3D reconstruction of an object by taking a series of 2D projections from a wide range of orientations. Ideally, projection images from À90 to þ90 around a fixed axis are needed to render a perfect 3D reconstruction. [19] However, due to the limited space in an electron microscope or shadowing from the holder/sample/ support, it is hard to obtain projections for the full tilt range in many conditions. With a specialized tomography holder, images from À70 to þ70 can be obtained in most TEMs, whereas in liquid-cell tomography, the tilt range is very limited to typically <30 , because the liquid flow holders are much bulkier. [20] The limited tilt range unavoidably leads to a "missing wedge" of information, which renders artifacts in the reconstructed tomograms. These artifacts introduced by the missing wedge reduce the resolution and reliability of the reconstructed tomograms and, sometimes, can even lead to serious misinterpretations. [6] So far, the missing-wedge problem has been the main source of systematic error that limits the application of electron tomography. Mathematically speaking, when there are insufficient number of projections, the inverse problem of tomography is ill-posed, because the solution is non-unique. Therefore, to constrain the solution space, strong priors need to be used to regularize the problem. For example, total variance minimization (TVM) combines iterative reconstruction and regularization of total variance to recover the lost information and reduces the artifacts introduced by the missing wedge. [21,22] This method is inspired by compressive sensing, [23,24] and it essentially deploys the sparsity constraint in the gradient domain of the tomogram. Some caveats of TVM are that it is not parameter-free and is also computationally expensive. Keeping these aside, the real problem of TVM or any generalized TVM approach is that they are bound to one regularization that promotes one prior constraint on the solution, which may or may not be suitable for the object of interest. For example, TVM promotes piecewise constants, rendering cartoon-like tomograms lacking gradient details.

show abstract

Knowledge-Based Multi-sequence MR Segmentation via Deep Learning with a Hybrid U-Net++ Model

Cited by 6 publications

References 19 publications

U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications

U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications

Effective extraction of ventricles and myocardium objects from cardiac magnetic resonance images with a multi-task learning U-Net

0.7 Å Resolution Electron Tomography Enabled by Deep‐Learning‐Aided Information Recovery

Contact Info

Product

Resources

About