IterNet: Retinal Image Segmentation Utilizing Structural Redundancy in Vessel Networks

Li, Liangzhi; Verma, Manisha; Nakashima, Yuta; Nagahara, Hajime; Kawasaki, Ryo

doi:10.1109/wacv45572.2020.9093621

Cited by 213 publications

(138 citation statements)

References 23 publications

Supporting

Mentioning

137

Contrasting

Unclassified

Order By: Relevance

“…We consider existing learning-based vessel detection methods for as baselines for comparison. These methods include HED [48], U-Net [13], DRIU [11], CRF [49],NestUNet [14], M2U-Net [50], CE-Net [18], CS-Net [51], RU-Net [15], and IterNet [16]. We train all methods on the IOSTAR [26] dataset where the images are captured with the scanning laser ophthalmoscopy (SLO) technique that is also used in the PRIME-FP20 dataset.…”

Section: Implementation Details and Alternative Methodsmentioning

confidence: 99%

“…Recently, supervised learning appproaches, specifically deep neural networks (DNNs), have led to significant improvements in retinal vessel detection. A variety of DNN architectures have been proposed for retinal vessel detection, including perpixel classifier [10], fully convolutional network [11], [12], U-Net [13]- [16], graph neural network [17], context encoder network [18], and generative adversarial networks [19]. Additionally, several works exploit novel loss functions [20]- [22] and training strategies [23].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Weakly-Supervised Vessel Detection in Ultra-Widefield Fundus Photography via Iterative Multi-Modal Registration and Learning

Ding

Kuriyan

Ramchandran

et al. 2021

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Section: Implementation Details and Alternative Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Weakly-Supervised Vessel Detection in Ultra-Widefield Fundus Photography via Iterative Multi-Modal Registration and Learning

Ding

Kuriyan

Ramchandran

et al. 2021

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

“…This information can be learned by the neural network carefully trained on well-annotated examples, where the composition of the vessels is known. This was implemented in the network SeqNet, 27 specially trained to distinguish between veins and arteries.…”

Section: Blood Vessels Analysismentioning

confidence: 99%

“…C) Fragment of the original image A. D) Overlay of predicted vessel structure and the original fragment image c. Arteries (red) and veins (blue) predicted by SeqNet 27 neural network; E) Segmented arteries and veins. Predicted arteries (red) and veins (blue) by SeqNet neural network.…”

Section: Blood Vessels Analysismentioning

confidence: 99%

Deep learning: step forward to high-resolution in vivo shortwave infrared imaging

Baulin¹,

Usson²,

Guével³

2021

Preprint

View full text Add to dashboard Cite

In vivo optical imaging is a fast growing field that offers great perspectives for biomedical applications. In particular, imaging in the shortwave infrared window (SWIR: 1000-1700 nm) represents major improvement compared to the NIR-I region (700-900 nm) in terms of temporal and spatial resolutions in depths down to 4 mm. SWIR is a fast and cheap alternative to more precise methods such as X-ray and opto-acoustic imaging. Main obstacles in SWIR imaging are the noise and scattering from tissues and skin that reduce the precision of the method. We demonstrate the combination of SWIR in vivo imaging in the NIR-IIb region (1500-1700 nm) with advanced deep learning image analysis allows to overcome these obstacles and making a large step forward to high resolution imaging: it allows to precisely segment vessels from tissues and noise, provides morphological structure of the vessels network, with learned pseudo-3D shape, their relative position, dynamic information of blood vascularization in depth in small animals and distinguish the vessels types: artieries and veins. For demonstration we use neural the network IterNet that exploits structural redundancy of the blood vessels (L. Li, et.al., The IEEE WACV, 2020), which provides a useful analysis tool for raw SWIR images.

show abstract

“…In [25], Zhang et al introduced an attention guided network (AG-Net) to achieve the retinal blood map. Li et al [26] designed a mini-UNets architecture performed based on the output of classical U-Net that further achieve the obscured detail of vessel.…”

Section: Related Workmentioning

confidence: 99%

Retinal vessel segmentation with constrained-based nonnegative matrix factorization and 3D modified Attention U-Net

Yao

Zhu

2020

Preprint

View full text Add to dashboard Cite

Due to the complex morphology and characteristic of retinal vessels, it remains challenging for most of the existing algorithms to accurately detect them. This paper proposes a supervised retinal vessels extraction scheme using constrained-based Nonnegative Matrix Factorization (NMF) and 3D modified Attention U-Net architecture. The proposed method detects the retinal vessels by three major steps. First, we perform Gaussian filter and Gamma correction on the Green channel of retinal images to suppress background noise and adjust the contrast of images. Then, the study develops a new within-class and between-class constrained NMF algorithms to extract neighborhood feature information of every pixel and reduce feature data dimension. By using these constraints, the method can effectively gather similar features within-class and discriminate features between-class to improve feature description ability for each pixel. Next, this study formulates segmentation task as a classification problem and solves it with a more contributing 3D modified Attention U-Net as a two-labels classifier for reducing computational cost. This proposed network contains an up-sampling to raise image resolution before encoding and revert image to its original size with a down-sampling after three max-pooling layers. Besides, the Attention Gate (AG) set in these layers contributes to more accurate segmentation by maintaining details while suppressing noises. Finally, the experimental results on two publicly available datasets STARE and DRIVE demonstrate better performance than most existing methods.

show abstract

IterNet: Retinal Image Segmentation Utilizing Structural Redundancy in Vessel Networks

Cited by 213 publications

References 23 publications

Weakly-Supervised Vessel Detection in Ultra-Widefield Fundus Photography via Iterative Multi-Modal Registration and Learning

Weakly-Supervised Vessel Detection in Ultra-Widefield Fundus Photography via Iterative Multi-Modal Registration and Learning

Deep learning: step forward to high-resolution in vivo shortwave infrared imaging

Retinal vessel segmentation with constrained-based nonnegative matrix factorization and 3D modified Attention U-Net

Contact Info

Product

Resources

About