Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 7, 1080(2025)
Unsupervised source-free multi-domain adaptive diabetic retinopathy classification
Fig. 1. Framework of the source-free multi-domain DR classification model based on Diff-DATL
Fig. 3. t-SNE visualization of the Diff-DATL model in RDR and NAD tasks with 1 000 samples randomly selected for each class
Fig. 4. Ablation study results of Diff-DATL model in RAR and NAD tasks under different balance parameters
Table 1. Comparative experimental results of the models under the RDR classification task
View tableView in ArticleTable 1. Comparative experimental results of the models under the RDR classification task
诊断模型 Accuracy Precision Sensitivity Specificity F1-score DANN 0.781 2 0.858 0 0.552 8 0.937 5 0.672 4 JAN 0.789 1 0.923 1 0.524 5 0.970 1 0.669 0 ADDA 0.852 5 0.845 4 0.779 4 0.902 5 0.811 1 CDAN 0.816 2 0.852 1 0.662 4 0.921 4 0.745 4 Diff-DATL 0.906 6 0.874 7 0.854 1 0.916 3 0.864 2
Table 2. Comparative experimental results of the models under the NAD task
View tableView in ArticleTable 2. Comparative experimental results of the models under the NAD task
诊断模型 Accuracy Precision Sensitivity Specificity F1-score DANN 0.964 2 0.965 2 0.927 3 0.965 7 0.945 9 JAN 0.905 7 0.955 4 0.854 1 0.959 0 0.901 7 ADDA 0.833 9 0.959 5 0.702 2 0.969 5 0.810 9 CDAN 0.790 8 0.924 5 0.639 7 0.946 3 0.756 2 Diff-DATL 0.967 5 0.992 3 0.904 7 0.992 7 0.946 5
Get Citation
Copy Citation Text
Guanghua ZHANG, Yang YANG, Guohua XU. Unsupervised source-free multi-domain adaptive diabetic retinopathy classification[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(7): 1080
Category:
Received: Feb. 6, 2025
Accepted: --
Published Online: Aug. 11, 2025
The Author Email: Guanghua ZHANG (zhangguanghua@tyu.edu.cn)
