Referable diabetic retinopathy identification from eye fundus images with weighted path for convolutional neural network
作者：
Highlights：
• A novel weighted path convolutional neural network (CNN) architecture, called WP-CNN, is proposed to classify the diabetic retinopathy and achieves an accuracy of 94.23% with sensitivity of 90.94%, specificity of 95.74%, an area under the receiver operating curve of 0.9823 and F1-score of 0.9087.
• The WP-CNN can be built by stacking weighted path blocks. The output of the weighted block can obtain the accurate diagnosis feature and reducing the multipath feature redundancy.
• Comparing with the state-of-art CNN architectures, WP-CNN can be trained faster and obtain the better classification performance with only one third of convolution layers number.
摘要
•A novel weighted path convolutional neural network (CNN) architecture, called WP-CNN, is proposed to classify the diabetic retinopathy and achieves an accuracy of 94.23% with sensitivity of 90.94%, specificity of 95.74%, an area under the receiver operating curve of 0.9823 and F1-score of 0.9087.•The WP-CNN can be built by stacking weighted path blocks. The output of the weighted block can obtain the accurate diagnosis feature and reducing the multipath feature redundancy.•Comparing with the state-of-art CNN architectures, WP-CNN can be trained faster and obtain the better classification performance with only one third of convolution layers number.
论文关键词：Diabetic retinopathy,Eye fundus images,Deep learning,Convolutional neural network
论文评审过程：Received 19 December 2018, Revised 25 June 2019, Accepted 9 July 2019, Available online 10 July 2019, Version of Record 12 August 2019.
论文官网地址：https://doi.org/10.1016/j.artmed.2019.07.002
