An Improved DenseNet Method Based on Transfer Learning for Fundus Medical Images

Abstract

There emerges an increasing need to improve the accuracy of computer recognition of fundus medical images. Remarkable progress has been made in image recognition, primarily due to the availability of large-scale annotated datasets and deep convolutional neural networks (CNNs). However, obtaining datasets as comprehensively annotated as ImageNet in the medical imaging domain remains a challenge. In this study, an improved DensenNet method based on Transfer Learning techniques is proposed for fundus medical images. Two experiments for fundus medical image data have been conducted respectively. The first one is to train the DenseNet models from scratch; the second one is fine-tuning operations by transfer learning, in which the DenseNet models pre-trained from natural image dataset to fundus medical images are improved. Experimental Results prove that the proposed method can improve the accuracy of fundus medical image classification, which is valuable for medical diagnosis.