Classification of Diabetic Retinopathy with Feature Fusion Network

Shuang Zhao; Ge Mu; Wenhua Zhao; Zhiqing Ma

doi:10.3788/LOP222415

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 14 >Page 1417001 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 14, 1417001 (2023)

Classification of Diabetic Retinopathy with Feature Fusion Network

Shuang Zhao¹, Ge Mu¹, Wenhua Zhao^2、*, and Zhiqing Ma²

Author Affiliations

¹Laboratory Management Office, Shandong University of Traditional Chinese Medicine, Jinan 250355, Shandong, China

²College of Intelligence and Information Engineering, Shandong University of Traditional Chinese Medicine, Jinan 250355, Shandong, China

show less

DOI: 10.3788/LOP222415 Cite this Article Set citation alerts

Shuang Zhao, Ge Mu, Wenhua Zhao, Zhiqing Ma. Classification of Diabetic Retinopathy with Feature Fusion Network[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1417001 Copy Citation Text

show less

Fig. 1. Five categories in the APTOS 2019 dataset. (a) No DR; (b) Mild DR; (c) Moderate DR; (d) Severe DR; (e) Proliferative DR

Download full size

Four types of images in the APTOS dataset. (a) Rectangular image, no cropping; (b) rectangular image, lossy vertical cropping; (c) square image, tight cropping; (d) rectangular image, lossy vertical and horizontal cropping

Fig. 2. Four types of images in the APTOS dataset. (a) Rectangular image, no cropping; (b) rectangular image, lossy vertical cropping; (c) square image, tight cropping; (d) rectangular image, lossy vertical and horizontal cropping

Download full size

Fig. 3. Preprocessed images

Download full size

Fig. 4. Overall network architecture

Download full size

Fig. 5. Multi-scale channel attention module

Download full size

Fig. 6. Loss curve

Download full size

Fig. 7. Confusion matrix

Download full size

Label	Number of images
No DR	1805
Mild DR	370
Moderate DR	999
Severe DR	193
Proliferative DR	295

Table 1. Distribution of APTOS 2019 dataset

Stage	Operateor	k×k	Output size	Number of layers
1	Conv	3×3	256×256×32	×1
2	MBConvBlock1	3×3	128×128×16	×1
3	MBConvBlock6	3×3	128×128×24	×2
4	MBConvBlock6	5×5	64×64×40	×2
5	MBConvBlock6	3×3	32×32×80	×3
6	MBConvBlock6	5×5	16×16×112	×3
7	MBConvBlock6	5×5	16×16×192	×4
8	MBConvBlock6	3×3	8×8×320	×1

Table 2. Network architecture of EfficientNet-B0 (stage1-8)

Model	Accuracy	Recall	Precision
EfficientNet-B0	83.92	62.33	78.94
EfficientNet-B0+Dilated convolution	84.74	64.68	77.08
Proposed model	85.25	68.41	82.20

Table 3. Ablation study

Model	Accuracy	Recall	Precision
ResNet-50^［34］	73.778
SE-Net^［34］	72.873
SE-ResNet-50^［34］	76.144
Model in Ref.［23］	80.06	80.00	81.00
Model in Ref.［25］	83.09
Proposed model	85.25	68.41	82.20

Table 4. Result comparison with existing researches

Shuang Zhao, Ge Mu, Wenhua Zhao, Zhiqing Ma. Classification of Diabetic Retinopathy with Feature Fusion Network[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1417001

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information