A large, open source dataset of stroke anatomical brain images and manual lesion segmentations

Liew, Sook‐Lei; Anglin, Julia; Banks, Nick W.; Sondag, Matt; Ito, Kaori; Kim, Ho‐Sung; Chan, Jennifer; Ito, Joyce; Jung, Connie; Khoshab, Nima; Lefebvre, Stéphanie; Nakamura, William; Saldaña, David; Schmiesing, Allie; Tran, Cathy; Vo, Danny; Ard, Tyler; Heydari, Panthea; Kim, Bokkyu; Aziz‐Zadeh, Lisa; Cramer, Steven C.; Liu, Jingchun; Soekadar, Surjo R.; Nordvik, Jan-Egil; Wang, Junping; Winstein, Carolee J.; Yu, Chunshui; Ai, Lei; Koo, Bonhwang; Craddock, R. Cameron; Milham, Michael P.; Lakich, Matthew; Pienta, Amy; Stroud, Alison

doi:10.1101/179614

Cited by 71 publications

(120 citation statements)

References 30 publications

Supporting

Mentioning

120

Contrasting

Order By: Relevance

“…We use FLAIR and T2, which have imminent visual features of tumors. Second, Anatomical Tracings of Lesions After Stroke (ATLAS) 4 [10] dataset is used for stroke MRI generation. The dataset contains 220 T1w images, which have diverse stroke lesions.…”

Section: Dataset and Preprocessingmentioning

confidence: 99%

Generation of 3D Brain MRI Using Auto-Encoding Generative Adversarial Networks

Kwon

Han

Kim

2019

Lecture Notes in Computer Science

101

102

View full text Add to dashboard Cite

As deep learning is showing unprecedented success in medical image analysis tasks, the lack of sufficient medical data is emerging as a critical problem. While recent attempts to solve the limited data problem using Generative Adversarial Networks (GAN) have been successful in generating realistic images with diversity, most of them are based on image-to-image translation and thus require extensive datasets from different domains. Here, we propose a novel model that can successfully generate 3D brain MRI data from random vectors by learning the data distribution. Our 3D GAN model solves both image blurriness and mode collapse problems by leveraging α-GAN that combines the advantages of Variational Auto-Encoder (VAE) and GAN with an additional code discriminator network. We also use the Wasserstein GAN with Gradient Penalty (WGAN-GP) loss to lower the training instability. To demonstrate the effectiveness of our model, we generate new images of normal brain MRI and show that our model outperforms baseline models in both quantitative and qualitative measurements. We also train the model to synthesize brain disorder MRI data to demonstrate the wide applicability of our model. Our results suggest that the proposed model can successfully generate various types and modalities of 3D whole brain volumes from a small set of training data.

show abstract

Section: Dataset and Preprocessingmentioning

confidence: 99%

Generation of 3D Brain MRI Using Auto-Encoding Generative Adversarial Networks

Kwon

Han

Kim

2019

Lecture Notes in Computer Science

101

102

View full text Add to dashboard Cite

show abstract

“…Acute lesions were on average smaller (our data: 12051 mm 3 ) than typically seen in chronic stroke (reported average chronic lesion sizes vary between 21280 and 34176 mm 3 ; cf. Corbetta et al, 2015;Liew et al, 2018), providing three distinct advantages for LBM analysis.…”

Section: Lesion Distributionmentioning

confidence: 99%

Dissociation between Frontal and Temporal-Parietal Contributions to Connected Speech in Acute Stroke

Ding

Martin

Hamilton

et al. 2019

Preprint

View full text Add to dashboard Cite

Humans are uniquely able to retrieve and combine words into syntactic structure to produce connected speech. Previous identification of focal brain regions necessary for production focused primarily on associations with the content produced by speakers with chronic stroke, where function may have shifted to other regions after reorganization occurred. Here, we relate patterns of brain damage with deficits to the content and structure of spontaneous connected speech in 52 speakers during the acute stage of a left hemisphere stroke. Multivariate lesion behavior mapping demonstrated that damage to temporal-parietal regions impacted the ability to retrieve words and produce them within increasingly complex combinations. Damage primarily to inferior frontal cortex affected the production of syntactically accurate structure. In contrast to previous work, functional-anatomical dissociations did not depend on lesion size likely because acute lesions were smaller than typically found in chronic stroke. These results are consistent with predictions from theoretical models based primarily on evidence from language comprehension and highlight the importance of investigating individual differences in brain-language relationships in speakers with acute stroke.

show abstract

“…In this section, we also show the superiority of the proposed loss and discuss the results of the proposed dimension-transform block at different stages. Datasets and quantitative indicators: We have used the Anatomical Tracings of Lesions-After-Stroke (ATLAS) dataset [7] as our training and validation sets. The dataset contains 229 cases of chronic stroke with MRI T1 sequence scans, in which the size of each case is 233×197×189 while the physical size is 0.9×0.9×3.0mm 3 ; the scans delineate different lesion grade staging.…”

Section: Experimental Results and Discus-sionsmentioning

confidence: 99%

D-UNet: A Dimension-Fusion U Shape Network for Chronic Stroke Lesion Segmentation

Zhou

Huang

Dong

et al. 2021

IEEE/ACM Trans. Comput. Biol. and Bioinf.

146

View full text Add to dashboard Cite

Assessing the location and extent of lesions caused by chronic stroke is critical for medical diagnosis, surgical planning, and prognosis. In recent years, with the rapid development of 2D and 3D convolutional neural networks (CNN), the encoder-decoder structure has shown great potential in the field of medical image segmentation. However, the 2D CNN ignores the 3D information of medical images, while the 3D CNN suffers from high computational resource demands. This paper proposes a new architecture called dimension-fusion-UNet (D-UNet), which combines 2D and 3D convolution innovatively in the encoding stage. The proposed architecture achieves a better segmentation performance than 2D networks, while requiring significantly less computation time in comparison to 3D networks. Furthermore, to alleviate the data imbalance issue between positive and negative samples for the network training, we propose a new loss function called Enhance Mixing Loss (EML). This function adds a weighted focal coefficient and combines two traditional loss functions. The proposed method has been tested on the ATLAS dataset and compared to three state-of-the-art methods. The results demonstrate that the proposed method achieves the best quality performance in terms of DSC = 0.5349±0.2763 and precision = 0.6331±0.295).

show abstract

A large, open source dataset of stroke anatomical brain images and manual lesion segmentations

Cited by 71 publications

References 30 publications

Generation of 3D Brain MRI Using Auto-Encoding Generative Adversarial Networks

Generation of 3D Brain MRI Using Auto-Encoding Generative Adversarial Networks

Dissociation between Frontal and Temporal-Parietal Contributions to Connected Speech in Acute Stroke

D-UNet: A Dimension-Fusion U Shape Network for Chronic Stroke Lesion Segmentation

Contact Info

Product

Resources

About