Research on Medical Image Classification Based on Image Segmentation and Feature Fusion

Abstract

Pneumonia has always been the leading infectious disease leading to the death of children under five years old. X-ray images of the lungs have become the key to the diagnosis of this disease. If computer-aided medical diagnosis is used to automatically detect lung abnormalities, the accuracy of the diagnosis will be improved.This article aims to introduce a deep learning technology based on the combination of image segmentation and feature fusion, which can realize automatic diagnosis of patients with pneumonia in X-ray images. First, use residuals to achieve image segmentation to more accurately divide the lung area. Secondly, the Xception network is used to extract the in-depth features of the data, and the extracted features are passed to the LSTM model to detect the extracted features, and classify the two cases of pneumonia and no pneumonia. This research combines Pearson's feature selection ideas and fuses the correlation between the two loss functions. Experimental results show that the accuracy of this paper is 98%, and the accuracy of AUC is 99%. Compared with the existing technical methods, the accuracy of the model designed in this paper is greatly improved. The model we designed has achieved excellent. experimental results on the currently available data sets. I hope our research can help doctors in the detection of pneumonia in children.