基于改进卷积神经网络的视网膜血管图像分割

吴晨玥; 易本顺; 章云港; 黄松; 冯雨

doi:doi:10.3788/AOS201838.1111004

光学学报, 2018, 38 (11): 1111004, 网络出版: 2019-05-09

基于改进卷积神经网络的视网膜血管图像分割下载： 1715次

Retinal Vessel Image Segmentation Based on Improved Convolutional Neural Network

吴晨玥 ^*易本顺 ^*章云港黄松冯雨

作者单位

武汉大学电子信息学院, 湖北武汉 430072

图 & 表

图 1. RDB示意图

Fig. 1. Schematic of RDB

下载图片查看原文

图 2. 不同扩张率下的空洞卷积示意图。(a) r=1; (b) r=2; (c) r=3

Fig. 2. Schematic of dilated convolutions under different dilation rates. (a) r=1; (b) r=2; (c) r=3

下载图片查看原文

图 3. 网络结构示意图

Fig. 3. Schematic of network structure

下载图片查看原文

图 4. 本文算法与文献[ 9]算法的分割效果。(a)源图像;(b)金标准图像;(c)文献[ 9]结果;(d)本文算法结果

Fig. 4. Segmentation results by proposed method and method in Ref. [9]. (a) Original images; (b) ground truth; (c) results in Ref. [9]; (d) results by proposed method

下载图片查看原文

图 5. 本文算法与文献[ 9]算法的分割结果ROC曲线。(a) ROC曲线;(b)图(a)中红框区域内曲线

Fig. 5. ROC curves of segmentation results by proposed method and method in Ref. [9]. (a) ROC curves; (b) curves in red box of Fig. (a)

下载图片查看原文

表 1不同算法在DRIVE数据集上视网膜血管分割性能对比

Table1. Retinal vessel segmentation performance comparison among different algorithms on DRIVE datasets

Method		Accuracy	Sensitivity	Specificity	AUC
Traditional method	Method in Ref. [4]	0.9489	0.7657	0.9809	-
	Method in Ref. [5]	0.9539	0.7039	0.9783	-
	Method in Ref. [6]	-	0.7897	0.9684	-
	Method in Ref. [22]	0.9457	0.7527	0.9744	-
Deep learning method	Method in Ref. [9]	0.9535	0.7811	0.9807	0.9765
	Method in Ref. [10]	0.9533	0.7464	0.9836	0.9752
	Proposed method	0.9556	0.8036	0.9778	0.9800

查看原文

表 2Retinal vessel segmentation performance comparison among different network structures on DRIVE datasets

Table2.

Method	Accuracy	Sensitivity	Specificity	AUC
Network_1	0.9537	0.7897	0.9777	0.9780
Network_2	0.9537	0.7968	0.9766	0.9772
Network_3	0.9542	0.7986	0.9768	0.9784
Network_4	0.9548	0.7973	0.9778	0.9792
Proposed	0.9556	0.8036	0.9778	0.9800

查看原文

吴晨玥, 易本顺, 章云港, 黄松, 冯雨. 基于改进卷积神经网络的视网膜血管图像分割[J]. 光学学报, 2018, 38(11): 1111004. Chenyue Wu, Benshun Yi, Yungang Zhang, Song Huang, Yu Feng. Retinal Vessel Image Segmentation Based on Improved Convolutional Neural Network[J]. Acta Optica Sinica, 2018, 38(11): 1111004.

基于改进卷积神经网络的视网膜血管图像分割下载： 1715次

图 1. RDB示意图

Fig. 1. Schematic of RDB

图 2. 不同扩张率下的空洞卷积示意图。(a) r=1; (b) r=2; (c) r=3

Fig. 2. Schematic of dilated convolutions under different dilation rates. (a) r=1; (b) r=2; (c) r=3

图 3. 网络结构示意图

Fig. 3. Schematic of network structure

图 4. 本文算法与文献[ 9]算法的分割效果。(a)源图像;(b)金标准图像;(c)文献[ 9]结果;(d)本文算法结果

Fig. 4. Segmentation results by proposed method and method in Ref. [9]. (a) Original images; (b) ground truth; (c) results in Ref. [9]; (d) results by proposed method

图 5. 本文算法与文献[ 9]算法的分割结果ROC曲线。(a) ROC曲线;(b)图(a)中红框区域内曲线

Fig. 5. ROC curves of segmentation results by proposed method and method in Ref. [9]. (a) ROC curves; (b) curves in red box of Fig. (a)

表 1不同算法在DRIVE数据集上视网膜血管分割性能对比

Table1. Retinal vessel segmentation performance comparison among different algorithms on DRIVE datasets

表 2Retinal vessel segmentation performance comparison among different network structures on DRIVE datasets

Table2.

关于本站 Cookie 的使用提示

全站搜索

基于改进卷积神经网络的视网膜血管图像分割 下载： 1715次

图 1. RDB示意图

Fig. 1. Schematic of RDB

图 2. 不同扩张率下的空洞卷积示意图。(a) r=1; (b) r=2; (c) r=3

Fig. 2. Schematic of dilated convolutions under different dilation rates. (a) r=1; (b) r=2; (c) r=3

图 3. 网络结构示意图

Fig. 3. Schematic of network structure

图 4. 本文算法与文献[ 9]算法的分割效果。(a)源图像;(b)金标准图像;(c)文献[ 9]结果;(d)本文算法结果

Fig. 4. Segmentation results by proposed method and method in Ref. [9]. (a) Original images; (b) ground truth; (c) results in Ref. [9]; (d) results by proposed method

图 5. 本文算法与文献[ 9]算法的分割结果ROC曲线。(a) ROC曲线;(b)图(a)中红框区域内曲线

Fig. 5. ROC curves of segmentation results by proposed method and method in Ref. [9]. (a) ROC curves; (b) curves in red box of Fig. (a)

表 1不同算法在DRIVE数据集上视网膜血管分割性能对比

Table1. Retinal vessel segmentation performance comparison among different algorithms on DRIVE datasets

表 2Retinal vessel segmentation performance comparison among different network structures on DRIVE datasets

Table2.

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

基于改进卷积神经网络的视网膜血管图像分割下载： 1715次