notesum.ai

Published at November 6

Cross Feature Fusion of Fundus Image and Generated Lesion Map for Referable Diabetic Retinopathy Classification

eess.IV

cs.AI

cs.CV

Released Date: November 6, 2024

Authors: Dahyun Mok¹, Junghyun Bum², Le Duc Tai¹, Hyunseung Choo¹

Aff.: ¹Dept. of Electrical and Computer Engineering, Sungkyunkwan University, Korea; ²College of Computing and Informatics, Sungkyunkwan University, Korea

Arxiv: http://arxiv.org/abs/2411.03618v1

Model	ACC	AUC	TPR	TNR
ResNet50[10]	87.8	88.1	67.2	93.1
Swin-T(Tiny)[26]	91.6	91.7	81.3	92.2
Swin-S(Small)[26]	91.7	91.8	81.7	93.1
Swin-B(Base)[26]	91.7	91.7	81.8	93.4
Swin-L(Large)[26]	91.9	91.9	81.7	91.7
ConvNeXt[27]	87.5	85.1	76.4	81.6
CrossViT[4]	88.0	85.6	76.8	83.2
CoATNet[24]	81.4	79.4	72.6	77.3
Ensemble of Plain and Robust model[21]	90.2	91.0	79.2	88.1
Lesion-Aware Contrastive Learning[16]	84.6	83.8	73.4	85.2
Ours (Proposed Method)	94.6	96.2	82.0	97.6