Enhancement of Perivascular Spaces Using Densely Connected Deep Convolutional Neural Network

Jung, Euijin; Chikontwe, Philip; Zong, Xiaopeng; Lin, Weili; Shen, Dinggang; Park, Sang Hyun

doi:10.1109/access.2019.2896911

Cited by 27 publications

(18 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It should be noted that the aim of these techniques was to improve the accuracy of the automated segmentation, but our study primarily aimed to propose a map that improves the visibility and detectability of the PVS, which can also make the visual scoring more accurate. Enhancement of PVS was also proposed using Haar transformation 19 and more recently using conventional neural network 73 , both on T2w images acquired using 7T MRI. Our approach focuses on enhancing the contrast of the PVS by combining T1w and T2w images and was optimized for 3T MRI, which is more accessible in comparison with 7T MRI.…”

Section: Discussionmentioning

confidence: 99%

Image processing approaches to enhance perivascular space visibility and quantification using MRI

Sepehrband

Barisano

Sheikh‐Bahaei

et al. 2019

Sci Rep

146

View full text Add to dashboard Cite

Imaging the perivascular spaces (PVS), also known as Virchow-Robin space, has significant clinical value, but there remains a need for neuroimaging techniques to improve mapping and quantification of the PVS. Current technique for PVS evaluation is a scoring system based on visual reading of visible PVS in regions of interest, and often limited to large caliber PVS. Enhancing the visibility of the PVS could support medical diagnosis and enable novel neuroscientific investigations. Increasing the MRI resolution is one approach to enhance the visibility of PVS but is limited by acquisition time and physical constraints. Alternatively, image processing approaches can be utilized to improve the contrast ratio between PVS and surrounding tissue. Here we combine T1- and T2-weighted images to enhance PVS contrast, intensifying the visibility of PVS. The Enhanced PVS Contrast (EPC) was achieved by combining T1- and T2-weighted images that were adaptively filtered to remove non-structured high-frequency spatial noise. EPC was evaluated on healthy young adults by presenting them to two expert readers and also through automated quantification. We found that EPC improves the conspicuity of the PVS and aid resolving a larger number of PVS. We also present a highly reliable automated PVS quantification approach, which was optimized using expert readings.

show abstract

Section: Discussionmentioning

confidence: 99%

Image processing approaches to enhance perivascular space visibility and quantification using MRI

Sepehrband

Barisano

Sheikh‐Bahaei

et al. 2019

Sci Rep

146

View full text Add to dashboard Cite

show abstract

“…First is the quality and type of the input image used for the PVS detection, such as the strength of the acquisition MR scanner and image resolution. Some studies used data acquired either at 1.5 T (Dubost et al, 2020; Gonzalez-Castro et al, 2017), 3T ((Ballerini et al, 2018; Boespflug et al, 2018; Sepehrband et al, 2019; Sudre et al, 2019) and our data) or 7T (Jung et al, 2019; Lian et al, 2018). Up to 3T, the data are most commonly acquired with approximately the same 1 mm 3 sampling size, whereas with 7T scanners, sampling is usually 8 times higher, providing a crucial advantage for detecting small DWM PVS.…”

Section: Discussionmentioning

confidence: 99%

“…Therefore, there is a strong need for unsupervised quantitative volumetric segmentation methods that could ideally identify every PVS in each individual as a 3-dimensional (3D) object in a perfectly reproducible manner. In fact, in the past few years, there have been several attempts at developing such automated PVS segmentation methods using broadly two different approaches, one based primarily on image processing (Ballerini et al, 2018;Boespflug et al, 2018;Gonzalez-Castro et al, 2017;Schwartz et al, 2019;Sepehrband et al, 2019;Wang et al, 2016;Zhang et al, 2017) and the other mainly based on deep learning (DL) (Dubost et al, 2020;Dubost et al, 2019;Jung et al, 2019;Lian et al, 2018;Sudre et al, 2019). The former approach is based on signal enhancement/noise reduction and/or specifically tailored morphological filters derived from the precise analysis of a few PVSs.…”

Section: Introductionmentioning

confidence: 99%

3D segmentation of perivascular spaces on T1-weighted 3 Tesla MR images with a convolutional autoencoder and a U-shaped neural network

Boutinaud

Tsuchida

Laurent

et al. 2020

Preprint

View full text Add to dashboard Cite

We implemented a deep learning (DL) algorithm for the 3-dimensional segmentation of perivascular spaces (PVSs) in deep white matter (DWM) and basal ganglia (BG). This algorithm is based on an autoencoder and a U-shaped network (U-net), and was trained and tested using T1-weighted magnetic resonance imaging (MRI) data from a large database of 1,832 healthy young adults. An important feature of this approach is the ability to learn from relatively sparse data, which gives the present algorithm a major advantage over other DL algorithms. Here, we trained the algorithm with 40 T1-weighted MRI datasets in which all “visible” PVSs were manually annotated by an experienced operator. After learning, performance was assessed using another set of 10 MRI scans from the same database in which PVSs were also traced by the same operator and were checked by consensus with another experienced operator. The Sorensen-Dice coefficients for PVS voxel detection in DWM (resp. BG) were 0.51 (resp. 0.66), and 0.64 (resp. 0.71) for PVS cluster detection (volume threshold of 0.5 within a range of 0 to 1). Dice values above 0.90 could be reached for detecting PVSs larger than 10 mm3 and 0.95 for PVSs larger than 15 mm3. We then applied the trained algorithm to the rest of the database (1,782 individuals). The individual PVS load provided by the algorithm showed a high agreement with a semi-quantitative visual rating done by an independent expert rater, both for DWM and for BG. Finally, we applied the trained algorithm to an age-matched sample from another MRI database acquired using a different scanner. We obtained a very similar distribution of PVS load, demonstrating the interoperability of this algorithm.

show abstract

“…Due to the above-mentioned advantages, DenseNets find a variety of applications in modern fields [4,12]. The DenseNet structure proposed in this paper is shown in Fig.…”

Section: The Local Connections Also Provide Implicit Supervisionmentioning

confidence: 99%

Generation of SAR Images Using Deep Learning

Bhamidipati

Srivatsa

Gowda³

et al. 2020

SN COMPUT. SCI.

View full text Add to dashboard Cite

Synthetic aperture radars (SAR) are a type of high-resolution radars that are used for terrestrial and airborne imaging. They use a moving source to simulate a large antenna aperture, thereby providing a high-quality map of the surroundings. They are popular due to their wide variety of applications in airborne naval and defense systems as well as topography and seismology. In defense applications, they are mainly used for reconnaissance and mapping of enemy targets such as armored vehicles, tanks, and runways. In recent times, there has been a heavy need for machine learning algorithms that are capable of automatically recognizing the targets detected by the SAR without human input. Such algorithms demand large-scale labeled datasets for training, validation and testing. Due to the nature of the SAR, acquisition and collection of these labeled datasets is quite a tedious and expensive task requiring a lot of man hours. Thus, alternate methods to produce new SAR image data are needed. In this paper, generative adversarial networks (GAN) are used to generate new spotlight SAR images from a limited pre-existing dataset. To quantitatively verify the effectiveness of the images produced by the GAN, a novel convolutional neural network (CNN) is devised and trained on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset images.

show abstract

Enhancement of Perivascular Spaces Using Densely Connected Deep Convolutional Neural Network

Cited by 27 publications

References 35 publications

Image processing approaches to enhance perivascular space visibility and quantification using MRI

Image processing approaches to enhance perivascular space visibility and quantification using MRI

3D segmentation of perivascular spaces on T1-weighted 3 Tesla MR images with a convolutional autoencoder and a U-shaped neural network

Generation of SAR Images Using Deep Learning

Contact Info

Product

Resources

About