Scale-space approximated convolutional neural networks for retinal vessel segmentation

Kyoung Jin Noh, Sang Jun Park, Soochahn Lee

Research output: Contribution to journalArticleResearchpeer-review

Abstract

Background and objective: Retinal fundus images are widely used to diagnose retinal diseases and can potentially be used for early diagnosis and prevention of chronic vascular diseases and diabetes. While various automatic retinal vessel segmentation methods using deep learning have been proposed, they are mostly based on common CNN structures developed for other tasks such as classification. Methods: We present a novel and simple multi-scale convolutional neural network (CNN) structure for retinal vessel segmentation. We first provide a theoretical analysis of existing multi-scale structures based on signal processing. In previous structures, multi-scale representations are achieved through downsampling by subsampling and decimation. By incorporating scale-space theory, we propose a simple yet effective multi-scale structure for CNNs using upsampling, which we term scale-space approximated CNN (SSANet). Based on further analysis of the effects of the SSA structure within a CNN, we also incorporate residual blocks, resulting in a multi-scale CNN that outperforms current state-of-the-art methods. Results: Quantitative evaluations are presented as the area-under-curve (AUC) of the receiver operating characteristic (ROC) curve and the precision-recall curve, as well as accuracy, for four publicly available datasets, namely DRIVE, STARE, CHASE_DB1, and HRF. For the CHASE_DB1 set, the SSANet achieves state-of-the-art AUC value of 0.9916 for the ROC curve. An ablative analysis is presented to analyze the contribution of different components of the SSANet to the performance improvement. Conclusions: The proposed retinal SSANet achieves state-of-the-art or comparable accuracy across publicly available datasets, especially improving segmentation for thin vessels, vessel junctions, and central vessel reflexes.

Original languageEnglish
Pages (from-to)237-246
Number of pages10
JournalComputer Methods and Programs in Biomedicine
Volume178
DOIs
StatePublished - 1 Sep 2019

Fingerprint

Retinal Vessels
Neural networks
ROC Curve
Delayed Emergence from Anesthesia
Area Under Curve
Retinal Diseases
Vascular Diseases
Reflex
Early Diagnosis
Chronic Disease
Learning
Medical problems
Signal processing
Datasets

Keywords

  • Convolutional neural networks
  • Multi-scale representation
  • Retinal vessel segmentation
  • Scale-space approximation

Cite this

@article{d13477f967fe46ddae756a4da6659f1d,
title = "Scale-space approximated convolutional neural networks for retinal vessel segmentation",
abstract = "Background and objective: Retinal fundus images are widely used to diagnose retinal diseases and can potentially be used for early diagnosis and prevention of chronic vascular diseases and diabetes. While various automatic retinal vessel segmentation methods using deep learning have been proposed, they are mostly based on common CNN structures developed for other tasks such as classification. Methods: We present a novel and simple multi-scale convolutional neural network (CNN) structure for retinal vessel segmentation. We first provide a theoretical analysis of existing multi-scale structures based on signal processing. In previous structures, multi-scale representations are achieved through downsampling by subsampling and decimation. By incorporating scale-space theory, we propose a simple yet effective multi-scale structure for CNNs using upsampling, which we term scale-space approximated CNN (SSANet). Based on further analysis of the effects of the SSA structure within a CNN, we also incorporate residual blocks, resulting in a multi-scale CNN that outperforms current state-of-the-art methods. Results: Quantitative evaluations are presented as the area-under-curve (AUC) of the receiver operating characteristic (ROC) curve and the precision-recall curve, as well as accuracy, for four publicly available datasets, namely DRIVE, STARE, CHASE_DB1, and HRF. For the CHASE_DB1 set, the SSANet achieves state-of-the-art AUC value of 0.9916 for the ROC curve. An ablative analysis is presented to analyze the contribution of different components of the SSANet to the performance improvement. Conclusions: The proposed retinal SSANet achieves state-of-the-art or comparable accuracy across publicly available datasets, especially improving segmentation for thin vessels, vessel junctions, and central vessel reflexes.",
keywords = "Convolutional neural networks, Multi-scale representation, Retinal vessel segmentation, Scale-space approximation",
author = "Noh, {Kyoung Jin} and Park, {Sang Jun} and Soochahn Lee",
year = "2019",
month = "9",
day = "1",
doi = "10.1016/j.cmpb.2019.06.030",
language = "English",
volume = "178",
pages = "237--246",
journal = "Computer Methods and Programs in Biomedicine",
issn = "0169-2607",
publisher = "Elsevier Ireland Ltd",

}

Scale-space approximated convolutional neural networks for retinal vessel segmentation. / Noh, Kyoung Jin; Park, Sang Jun; Lee, Soochahn.

In: Computer Methods and Programs in Biomedicine, Vol. 178, 01.09.2019, p. 237-246.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Scale-space approximated convolutional neural networks for retinal vessel segmentation

AU - Noh, Kyoung Jin

AU - Park, Sang Jun

AU - Lee, Soochahn

PY - 2019/9/1

Y1 - 2019/9/1

N2 - Background and objective: Retinal fundus images are widely used to diagnose retinal diseases and can potentially be used for early diagnosis and prevention of chronic vascular diseases and diabetes. While various automatic retinal vessel segmentation methods using deep learning have been proposed, they are mostly based on common CNN structures developed for other tasks such as classification. Methods: We present a novel and simple multi-scale convolutional neural network (CNN) structure for retinal vessel segmentation. We first provide a theoretical analysis of existing multi-scale structures based on signal processing. In previous structures, multi-scale representations are achieved through downsampling by subsampling and decimation. By incorporating scale-space theory, we propose a simple yet effective multi-scale structure for CNNs using upsampling, which we term scale-space approximated CNN (SSANet). Based on further analysis of the effects of the SSA structure within a CNN, we also incorporate residual blocks, resulting in a multi-scale CNN that outperforms current state-of-the-art methods. Results: Quantitative evaluations are presented as the area-under-curve (AUC) of the receiver operating characteristic (ROC) curve and the precision-recall curve, as well as accuracy, for four publicly available datasets, namely DRIVE, STARE, CHASE_DB1, and HRF. For the CHASE_DB1 set, the SSANet achieves state-of-the-art AUC value of 0.9916 for the ROC curve. An ablative analysis is presented to analyze the contribution of different components of the SSANet to the performance improvement. Conclusions: The proposed retinal SSANet achieves state-of-the-art or comparable accuracy across publicly available datasets, especially improving segmentation for thin vessels, vessel junctions, and central vessel reflexes.

AB - Background and objective: Retinal fundus images are widely used to diagnose retinal diseases and can potentially be used for early diagnosis and prevention of chronic vascular diseases and diabetes. While various automatic retinal vessel segmentation methods using deep learning have been proposed, they are mostly based on common CNN structures developed for other tasks such as classification. Methods: We present a novel and simple multi-scale convolutional neural network (CNN) structure for retinal vessel segmentation. We first provide a theoretical analysis of existing multi-scale structures based on signal processing. In previous structures, multi-scale representations are achieved through downsampling by subsampling and decimation. By incorporating scale-space theory, we propose a simple yet effective multi-scale structure for CNNs using upsampling, which we term scale-space approximated CNN (SSANet). Based on further analysis of the effects of the SSA structure within a CNN, we also incorporate residual blocks, resulting in a multi-scale CNN that outperforms current state-of-the-art methods. Results: Quantitative evaluations are presented as the area-under-curve (AUC) of the receiver operating characteristic (ROC) curve and the precision-recall curve, as well as accuracy, for four publicly available datasets, namely DRIVE, STARE, CHASE_DB1, and HRF. For the CHASE_DB1 set, the SSANet achieves state-of-the-art AUC value of 0.9916 for the ROC curve. An ablative analysis is presented to analyze the contribution of different components of the SSANet to the performance improvement. Conclusions: The proposed retinal SSANet achieves state-of-the-art or comparable accuracy across publicly available datasets, especially improving segmentation for thin vessels, vessel junctions, and central vessel reflexes.

KW - Convolutional neural networks

KW - Multi-scale representation

KW - Retinal vessel segmentation

KW - Scale-space approximation

UR - http://www.scopus.com/inward/record.url?scp=85068535355&partnerID=8YFLogxK

U2 - 10.1016/j.cmpb.2019.06.030

DO - 10.1016/j.cmpb.2019.06.030

M3 - Article

VL - 178

SP - 237

EP - 246

JO - Computer Methods and Programs in Biomedicine

JF - Computer Methods and Programs in Biomedicine

SN - 0169-2607

ER -