Opto-Electronic Engineering, Volume. 52, Issue 6, 250082(2025)
Multi-frequency Transformer-guided graph-based feature aggregation for retinal image quality grading
Figures & Tables(17)
Fig. 5. Various steps of CLAHE and Gamma correction for retinal fundus images
Fig. 7. Feature heatmap of ablation experiments on Eye-Quality dataset
Fig. 8. Display of the functions of the FCRT module spectrum diagrams
Fig. 9. Network feature heatmaps of each stage on the Eye-Quality and RIQA-RFMiD datasets
Table 1. The number of class distribution of Eye-Quality datasets
View tableView in ArticleTable 1. The number of class distribution of Eye-Quality datasets
Eye-Quality 优质 可用 拒绝 总体 训练集 8347 1876 2320 12543 测试集 8470 4559 3220 16249 总体 16817 6435 5540 28792
Table 2. The number of class distribution of RIQA-RFMiD datasets
View tableView in ArticleTable 2. The number of class distribution of RIQA-RFMiD datasets
RIQA-RFMiD 优质 可用 拒绝 总体 训练集 1745 99 76 1920 测试集 494 95 51 640 总体 2239 194 127 2560
Table 3. Performance indicators of channel coefficient models at different frequencies on the Eye-Quality dataset
View tableView in ArticleTable 3. Performance indicators of channel coefficient models at different frequencies on the Eye-Quality dataset
Accuracy/% Precision/% Recall/% F1-Score/% 0.1 85.81 84.81 82.68 83.31 0.2 86.11 84.47 84.87 84.54 0.3 86.83 86.22 84.90 84.82 0.4 87.52 86.88 86.11 85.77 0.5 87.64 87.24 86.65 86.70 0.6 88.71 87.78 86.86 86.89 0.7 88.11 86.98 86.74 86.44 0.8 87.98 86.52 86.15 86.37 0.9 87.72 86.23 86.48 86.10
Table 4. Performance indicators of models with different weighting factors on the Eye-Quality dataset
View tableView in ArticleTable 4. Performance indicators of models with different weighting factors on the Eye-Quality dataset
Accuracy/% Precision/% Recall/% F1-Score/% 0 87.94 86.87 85.20 85.94 0.1 87.25 86.02 85.22 85.36 0.2 87.33 85.72 85.97 85.17 0.3 87.25 86.03 85.46 85.12 0.4 87.67 87.63 85.65 85.63 0.5 88.26 87.79 85.58 86.44 0.6 88.61 88.22 85.63 86.28 0.7 88.52 88.24 85.60 86.42 0.8 88.71 87.78 86.86 86.89 0.9 88.45 88.36 85.08 86.48 1.0 87.70 87.41 86.14 86.22
Table 5. Results of different algorithms on the Eye-Quality dataset
View tableView in ArticleTable 5. Results of different algorithms on the Eye-Quality dataset
Methods Accuracy/% Precision/% Recall/% F1-Score/% ResNeSt[ 22 ]87.35 85.63 84.82 85.06 NBIQA[ 23 ]79.17 76.41 75.09 74.41 FASNB[ 24 ]88.59 87.56 86.10 86.74 FBSM[ 25 ]87.23 86.40 84.54 85.13 DenseNet121-RGB [ 10 ]85.68 84.81 82.39 83.15 DenseNet121-MCF [ 10 ]87.22 85.63 84.82 85.06 Single-QA[ 13 ]88.47 87.15 86.45 86.62 MFTGA 88.71 87.78 86.86 86.89
Table 6. Results of different algorithms on the Eye-Quality dataset
View tableView in ArticleTable 6. Results of different algorithms on the Eye-Quality dataset
Methods Accuracy/% Precision/% Recall/% F1-Score/% FBSM[ 25 ]83.75 77.84 56.49 61.84 DenseNet121-RGB [ 10 ]83.13 71.87 57.41 60.42 DenseNet121-MCF [ 10 ]82.81 75.13 61.31 65.53 FASNB[ 24 ]84.22 73.95 62.94 67.17 MFTGA 84.95 74.22 68.25 67.58
Table 7. Ablation results on the Eye-Quality dataset
View tableView in ArticleTable 7. Ablation results on the Eye-Quality dataset
Methods Accuracy/% Precision/% Recall/% F1-Score/% W1 88.21 87.34 85.69 86.39 W2 87.01 85.81 84.26 84.68 W3 87.59 86.17 85.14 85.49 W4 88.18 88.11 84.71 86.17 W5 87.94 86.87 85.20 85.94 W6 88.71 87.78 86.86 86.89
Table 8. Correlation research experiment of MFTGA on the Eye-Quality dataset
View tableView in ArticleTable 8. Correlation research experiment of MFTGA on the Eye-Quality dataset
Methods Accuracy/% Precision/% Recall/% F1-Score/% M1 85.44 84.13 84.27 83.78 M2 86.58 85.48 85.03 84.69 M3 87.97 86.98 86.45 85.66 M4 88.04 87.50 86.70 86.25 M5 88.71 87.78 86.86 86.89
Tools
Get Citation
Copy Citation Text
Liming Liang, Yi Zhong, Chengbin Wang, Ting Kang. Multi-frequency Transformer-guided graph-based feature aggregation for retinal image quality grading[J]. Opto-Electronic Engineering, 2025, 52(6): 250082
Paper Information
Category: Article
Received: Mar. 15, 2025
Accepted: May. 8, 2025
Published Online: Sep. 3, 2025
The Author Email: Yi Zhong (钟奕)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20