Atlas Construction via Dictionary Learning and Group Sparsity

Shi, Feng; Wang, Li; Wu, Guorong; Yu, Zhengtao; Liu, Manhua; Gilmore, John H.; Lin, Weili; Shen, Dinggang

doi:10.1007/978-3-642-33415-3_31

Cited by 5 publications

(5 citation statements)

References 12 publications

(12 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bias correction was then performed on all images with N3 method [Sled et al, 1998] to reduce the impact of intensity inhomogeneity and thus improve the performance of the subsequent image processing. Non-brain tissues such as skull and dura were stripped with a learning-based meta algorithm [Shi et al, 2012]. Finally, brain tissues were segmented into gray matter (GM), white matter (WM), and ventricular cerebrospinal fluid (CSF) using a coupled level-set algorithm [Wang et al, 2011].…”

Section: Resultsmentioning

confidence: 99%

“…All image were preprocessed with the standard procedure which includes resampling, bias correction, skull stripping [Shi et al, 2012], and tissue segmentation [Wang et al, 2011]. The goal of image registration is to spatially normalize all preprocessed subject images into a common space, which is a necessary initial step for subsequent atlas building.…”

Section: Methodsmentioning

confidence: 99%

“…In our implementation, an atlas is constructed locally in a patch-by-patch fashion to ensure the local representativeness, and the neighboring patches are constrained to have similar representations by using the group sparsity strategy. A preliminary version of the method has been presented in a conference [Shi et al, 2012]. Here we extend it to include more descriptions of the method, novel anatomical features to further improve the performance, as well as extensive experiments to evaluate the proposed method.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Neonatal atlas construction using sparse representation

Shi

Wang

et al. 2014

Human Brain Mapping

Self Cite

View full text Add to dashboard Cite

Atlas construction generally includes first an image registration step to normalize all images into a common space and then an atlas building step to fuse the information from all the aligned images. Although numerous atlas construction studies have been performed to improve the accuracy of the image registration step, unweighted or simply weighted average is often used in the atlas building step. In this article, we propose a novel patch-based sparse representation method for atlas construction after all images have been registered into the common space. By taking advantage of local sparse representation, more anatomical details can be recovered in the built atlas. To make the anatomical structures spatially smooth in the atlas, the anatomical feature constraints on group structure of representations and also the overlapping of neighboring patches are imposed to ensure the anatomical consistency between neighboring patches. The proposed method has been applied to 73 neonatal MR images with poor spatial resolution and low tissue contrast, for constructing a neonatal brain atlas with sharp anatomical details. Experimental results demonstrate that the proposed method can significantly enhance the quality of the constructed atlas by discovering more anatomical details especially in the highly convoluted cortical regions. The resulting atlas demonstrates superior performance of our atlas when applied to spatially normalizing three different neonatal datasets, compared with other start-of-the-art neonatal brain atlases.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Neonatal atlas construction using sparse representation

Shi

Wang

et al. 2014

Human Brain Mapping

Self Cite

View full text Add to dashboard Cite

show abstract

“…It has been shown that a sparse signal can be recovered from a small number of its linear measurements with high probability 2,8 . These sparsity priors are employed in many computer vision applications, such as, but not limited to, robust face recognition 38 , image restoration 26 , image bias estimation 56 , MR reconstruction 24,16 , atlas construction 34 , resolution enhancement 54 , image bias estimation 57,56 and automatic image annotation 47,13 . Sparsity methods have been proved to be very effective at handling gross errors or outliers.…”

Section: Relevant Workmentioning

confidence: 99%

Adaptive Shape Prior Modeling via Online Dictionary Learning

Zhang

Zhan

Zhou

et al. 2013

Computer Vision in Medical Imaging

View full text Add to dashboard Cite

Shape" and "appearance", the two pillars of a deformable model, complement each other in object segmentation. In many medical imaging applications, while the low-level appearance information is weak or misleading , shape priors play a more important role to guide a correct segmentation. The recently proposed Sparse Shape Composition (SSC) 49,51 opens a new avenue for shape prior modeling. Instead of assuming any parametric model of shape statistics, SSC incorporates shape priors on-the-fly by approximating a shape instance (usually derived from appearance cues) by a sparse combination of shapes in a training repository. Theoretically, one can increase the modeling capability of SSC by including as many training shapes in the repository. However, this strategy confronts two limitations in practice. First, since SSC involves an iterative sparse optimization at run-time, the more shape instances contained in the repository, the less run-time efficiency SSC has. Therefore, a compact and informative shape dictionary is preferred to a large shape repository. Second, in medical imaging applications, training shapes seldom come in one batch. It is very time consuming and sometimes infeasible to reconstruct the shape dictionary every time new training shapes appear. In this chapter, we introduce an online learning method to address these two limitations. Our method starts from constructing an initial shape dictionary using the K-SVD algorithm. When new training shapes come, instead of reconstructing the dictionary from the ground up, we update the existing one using a block-coordinates descent approach. Using the dynamically updated dictionary, sparse shape composition can be gracefully scaled up to model shape priors from a large number of training shapes without sacrificing run-time efficiency. Our method is validated on lung localization in X-Ray and cardiac segmentation in MRI time series. Compared to the original SSC, it shows comparable performance while being significantly more efficient.

show abstract

“…Other methods 28,29 have focused on iteratively averaging the invertible elastic transformation of the template to better fit a target population. Additionally, template-free techniques 30,31 register all images simultaneously in a groupwise fashion; however, this is known to be computationally challenging due the size of the search space.…”

Section: Previous Workmentioning

confidence: 99%

Prostatome: A combined anatomical and disease based MRI atlas of the prostate

Rusu¹,

Bloch²,

Jaffe³

et al. 2014

Med. Phys.

View full text Add to dashboard Cite

Purpose:In this work, the authors introduce a novel framework, the anatomically constrained registration (AnCoR) scheme and apply it to create a fused anatomic-disease atlas of the prostate which the authors refer to as the prostatome. The prostatome combines a MRI based anatomic and a histology based disease atlas. Statistical imaging atlases allow for the integration of information across multiple scales and imaging modalities into a single canonical representation, in turn enabling a fused anatomical-disease representation which may facilitate the characterization of disease appearance relative to anatomic structures. While statistical atlases have been extensively developed and studied for the brain, approaches that have attempted to combine pathology and imaging data for study of prostate pathology are not extant. This works seeks to address this gap. Methods: The AnCoR framework optimizes a scoring function composed of two surface (prostate and central gland) misalignment measures and one intensity-based similarity term. This ensures the correct mapping of anatomic regions into the atlas, even when regional MRI intensities are inconsistent or highly variable between subjects. The framework allows for creation of an anatomic imaging and a disease atlas, while enabling their fusion into the anatomic imaging-disease atlas. The atlas presented here was constructed using 83 subjects with biopsy confirmed cancer who had pre-operative MRI (collected at two institutions) followed by radical prostatectomy. The imaging atlas results from mapping the in vivo MRI into the canonical space, while the anatomic regions serve as domain constraints. Elastic co-registration MRI and corresponding ex vivo histology provides "ground truth" mapping of cancer extent on in vivo imaging for 23 subjects. Results: AnCoR was evaluated relative to alternative construction strategies that use either MRI intensities or the prostate surface alone for registration. The AnCoR framework yielded a central gland Dice similarity coefficient (DSC) of 90%, and prostate DSC of 88%, while the misalignment of the urethra and verumontanum was found to be 3.45 mm, and 4.73 mm, respectively, which were measured to be significantly smaller compared to the alternative strategies. As might have been anticipated from our limited cohort of biopsy confirmed cancers, the disease atlas showed that most of the tumor extent was limited to the peripheral zone. Moreover, central gland tumors were typically larger in size, possibly because they are only discernible at a much later stage. Conclusions:The authors presented the AnCoR framework to explicitly model anatomic constraints for the construction of a fused anatomic imaging-disease atlas. The framework was applied to constructing a preliminary version of an anatomic-disease atlas of the prostate, the prostatome. The prostatome could facilitate the quantitative characterization of gland morphology and imaging features of prostate cancer. These techniques, may be applied on a large sample size data set to create...

show abstract

Atlas Construction via Dictionary Learning and Group Sparsity

Cited by 5 publications

References 12 publications

Neonatal atlas construction using sparse representation

Neonatal atlas construction using sparse representation

Adaptive Shape Prior Modeling via Online Dictionary Learning

Prostatome: A combined anatomical and disease based MRI atlas of the prostate

Contact Info

Product

Resources

About