基于GLCM和深度神经网络特征融合的小样本肺炎诊断算法
A Small Sample Pneumonia Diagnosis Algorithm Based on Feature Fusion of GLCM and Deep Neural Network

针对图像分类的各种深度神经网络模型发展迅速，但是在训练样本极少的情况下，深度学习模型普遍无法取得良好的样本外表现。纹理是医学影像中图像识别的关键特征，本文使用灰度共生矩阵(Gray Level Co-Occurrence Matrix, GLCM)提取胸部X光片的纹理特征，与预训练的ResNet50提取的特征进行融合，最后使用机器学习分类模型对胸部X光片进行肺炎识别。在本文的7种分类算法下，使用逻辑回归分类取得了最好的结果，该模型仅使用灰度共生矩阵提取的特征的分类准确率为80%，使用预训练的ResNet50提取的特征的分类准确率为90%，融合特征分类准确率为92.5%。实验结果显示，与仅使用一种方法提取特征的分类结果相比，融合特征再分类的方法使准确率有明显的提高。
Various deep neural network models for image classification have developed rapidly, but deep learning models generally cannot achieve good out-of-sample performance with very few training samples. Texture is a key feature of medical images in image recognition, we use GLCM to extract texture features from chest X-rays, which are fused with the features extracted from pretrained ResNet50. Finally, a machine learning classification model is used to recognize pneumonia from chest X-rays. Among the seven classification algorithms in this article, the best results were achieved using Logistic Regression. The classification accuracy of the model is 80% for features extracted using only the GLCM, 90% for features extracted using pre-trained ResNet50, and 92.5% for fused features. The experimental results show that compared to the classification results using only one method to extract features, the fusion feature re-classification method significantly improves the accuracy.