Abstract
The automatic diagnosis of various retinal diseases based on fundus images is important in supporting clinical decision-making. Convolutional neural networks (CNNs) have achieved remarkable results in such tasks. However, their high expression ability possibly leads to overfitting. Therefore, data augmentation (DA) techniques have been proposed to prevent overfitting while enriching datasets. Recent CNN architectures with more parameters render traditional DA techniques insufficient. In this study, we proposed a new DA strategy based on multimodal fusion (DAMF) which could integrate the standard DA method, data disrupting method, data mixing method, and autoadjustment method to enhance the image data in the training dataset to create new training images. In addition, we fused the results of the classifier by voting on the basis of DAMF, which further improved the generalization ability of the model. The experimental results showed that the optimal DA mode could be matched to the image dataset through our DA strategy. We evaluated DAMF on the iChallenge-PM dataset. At last, we compared training results between 12 DAMF processed datasets and the original training dataset. Compared with the original dataset, the optimal DAMF achieved an accuracy increase of 2.85% on iChallenge-PM.
Highlights
Pathologic myopia (PM) is one of the major causes of visual impairment worldwide [1,2,3]
The rectified linear unit (ReLU) nonlinear activation function was used to speed up the training of the network, multiGPU convolutional operations were implemented to address the limitations of insufficient graphic card resources at the time, and the DropOut random inactivation strategy was introduced to reduce overfitting at the full connection layer
DA strategy based on multimodal fusion (DAMF) strategies were implemented on the iChallenge-PM dataset, resulting in the formation of datasets, including the original one. en, the experiment used VGG-16 as a dataset picker to train each of these 13 datasets for 30 epochs, each epoch covering all the data once
Summary
Pathologic myopia (PM) is one of the major causes of visual impairment worldwide [1,2,3]. The model’s lack of expressive ability will lead to the weakness in the recognition of some rare lesion images [32] To address this problem, researchers have proposed optimized neural network models from different perspectives and achieved effective results. When the researchers optimize the models (such as widening and deepening models), they cannot predict the effectiveness of the models but just observe whether the optimization operation improves the performance of the original models through the training results Even if such optimization is effective, it may be computational and time-intensive or needs a long development cycle, so it cannot address the problem effectively [33,34,35]. The accuracy of the model in recognizing complex and rare case images will be effectively improved
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
