Fundus Image Screening for Diabetic Retinopathy

Jiayu Li; Minghui Chen; Ruijun Yang; Wenfei Ma; Xiangling Lai; Duowen Huang; Duxin Liu; Xinhong Ma; Yue Shen

doi:10.3788/CJL202249.1107001

Chinese Journal of Lasers, Volume. 49, Issue 11, 1107001(2022)

Fundus Image Screening for Diabetic Retinopathy

Jiayu Li¹, Minghui Chen^1、*, Ruijun Yang², Wenfei Ma¹, Xiangling Lai¹, Duowen Huang¹, Duxin Liu¹, Xinhong Ma¹, and Yue Shen¹

Author Affiliations

¹Shanghai Engineering Research Center of Interventional Medical, the Ministry of Education of Medical Optical Engineering Center, Department of Biomedical Engineering, School of Medical Instrument and Food Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

²The Third People’s Hospital of Mianyang City, Mianyang 621000, Sichuan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Architecture of CA-RepVGG

Download full size

Fig. 2. Ultra lightweight attention module ECA-Net

Download full size

Fig. 3. RepVGG module. (a) Training; (b) inference

Download full size

Fig. 4. Schematic of convolution transformation

Download full size

Fig. 5. Data enhancement. (a) Original image; (b) rotation of 60°；(c) mirror horizontally; (d) enhancement of brightness; (e) Gamma transformation; (f) adaptive histogram equalization

Download full size

Fig. 6. Confusion matrix of dataset 1

Download full size

Fig. 7. Confusion matrix of dataset 2

Download full size

Table 1. 2015 diabetic retinopathy detection dataset
View table
Table 1. 2015 diabetic retinopathy detection dataset
Grade Degree of illness Number of training images Number of testing images Total
0 Healthy 25810 39533 65343
1 Light 2443 3762 6205
2 Moderate 5292 7861 13153
3 Severe 873 1214 2087
4 Proliferative 708 1206 1914

Table 2. Comparison of performance indicators of different models for dataset 1

View table

Table 2. Comparison of performance indicators of different models for dataset 1

Model	Parameter number /10⁶	Speed /(photo·s^－1)	Accuracy /%	Precision /%	Sensitivity /%
VGG-16	13.84	208	82.0	83.9	85.0
Inception-V3	27.16	295	87.3	89.4	93.2
ResNet-50	25.56	360	89.5	90.7	97.0
ResNext-50	25.03	242	91.0	91.3	96.5
CA-RepVGG (ours)	51.82	415	92.4	91.6	96.5

Table 3. Comparison of performance indicators of different models for dataset 2

View table

Table 3. Comparison of performance indicators of different models for dataset 2

Model	Parameter number /10⁶	Speed /(photo·s^－1)	Accuracy /%	Precision /%	Sensitivity /%
VGG-16	13.84	208	83.0	82.7	84.0
Inception-V3	27.16	295	85.3	88.4	86.2
ResNet-50	25.56	360	89.4	89.7	91.0
ResNext-50	25.03	242	91.5	90.5	90.4
CA-RepVGG (ours)	51.82	415	93.9	96.3	93.8

Tools

Get Citation

Copy Citation Text

Jiayu Li, Minghui Chen, Ruijun Yang, Wenfei Ma, Xiangling Lai, Duowen Huang, Duxin Liu, Xinhong Ma, Yue Shen. Fundus Image Screening for Diabetic Retinopathy[J]. Chinese Journal of Lasers, 2022, 49(11): 1107001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites