一种用于黄斑病变分类的改进卷积神经网络模型

An Improved Convolutional Neural Network Modelfor Macular Diseases Classification

基于卷积神经网络的视网膜黄斑病变自动识别技术可辅助眼科医生诊断黄斑病变。为解决黄斑病变区域小和特征不明显导致黄斑病变类型不易识别的问题,提出了一种用于黄斑病变分类的改进卷积神经网络模型。首先,加入多尺度特征融合模块,将带有不同感受野的特征图进行拼接,从而提取更加丰富的黄斑病变特征;其次,增加注意力机制,有效抑制冗余特征的同时增加对病变区域的关注;最后,引入有效样本加权损失函数,充分学习少样本类别的病变特征,从而解决数据样本不平衡问题。实验证明,在UCSD视网膜黄斑病变数据集上,提出的模型进一步提高了黄斑病变的分类效果,分类准确率达到了97.60%,能够更加有效地辅助眼科医生诊断黄斑病变,提高诊疗效率。

Automatic recognition of retina macular diseases based on convolutional neural network can assist ophthalmologists in diagnosing macular diseases.An improved convolutional neural network model for macular diseases classification is proposed to solve the difficult problem of identifying the type of macular diseases caused by the small area and insignificant characteristics.Firstly,the multi-scale feature fusion module is added to splice the feature maps with different receptive fields and extract more abundant features of macular diseases;Secondly,the attention mechanism is embedded to effectively suppress redundant features and increase attention to the diseases area;Finally,the weighted loss based on effective number of samples is introduced to learn the pathological features of small sample categories for solving the problem of data sample imbalance.The experiment proves that the proposed model has further improved the classification effect of macular diseases on the UCSD dataset,and the classification accuracy rate has reached 97.60%.Therefore,the model improves the diagnosis and treatment efficiency by assisting ophthalmologists more effectively in diagnosing macular disease.