Fundus Image Generation and Classification of Diabetic Retinopathy Based on Convolutional Neural Network

• DOI: 10.3390/electronics13183603
• Publication Date: 2024-09-11T09:45:12Z
• Authors: Peiming Zhang; Jie Zhao; Qiaohong Liu; Xiao Liu; Xinyu Li; Yimeng Gao; Weiqi Li
• Keywords: Fundus (uterus)

To detect fundus diseases, for instance, diabetic retinopathy (DR) at an early stage, thereby providing timely intervention and treatment, a new grading method based on convolutional neural network is proposed. First, data cleaning enhancement are conducted to improve the image quality reduce unnecessary interference. Second, conditional generative adversarial with self-attention mechanism named SACGAN proposed augment number of images, addressing problems insufficient imbalanced samples. Next, improved DRMC Net, which combines ResNeXt-50 channel attention multi-branch residual module, classify retinopathy. Finally, gradient-weighted class activation mapping (Grad-CAM) utilized prove model’s interpretability. The outcomes experiment illustrates that has high accuracy, specificity, sensitivity, specific results 92.3%, 92.5%, respectively.