基于多模型比较的肺炎X射线影像识别方法研究

Research on Pneumonia X-Ray Image Recognition Method Based on Multi-Model Comparison

摘要: 目的：比较不同卷积神经网络模型在肺炎X射线影像识别任务中的性能，为医学影像智能识别模型的选择提供实验依据。方法：以公开肺炎胸部X射线影像数据集(Chest X-Ray Images (Pneumonia))为研究对象，选取AlexNet、VGG16、ResNet50、DenseNet121和MobileNetV2五种典型CNN模型进行对比实验，采用迁移学习策略进行模型训练，并使用Accuracy、Precision、Recall、F1-score和AUC等指标综合评价模型性能。结果：VGG16模型在肺炎识别任务中表现最优，Accuracy达90.38%，F1-score达92.59%。结论：不同CNN模型在肺炎X射线影像识别任务中存在显著性能差异，为临床辅助诊断系统的模型选择提供了参考依据。

Abstract: Objective: To compare the performance of different convolutional neural network (CNN) models in pneumonia X-ray image recognition tasks, and to provide experimental evidence for the selection of intelligent recognition models in medical imaging. Methods: Using the publicly available Chest X-Ray Images (Pneumonia) dataset, five typical CNN models—AlexNet, VGG16, ResNet50, DenseNet121, and MobileNetV2—were selected for comparative experiments. Transfer learning was adopted for model training, and model performance was comprehensively evaluated using Accuracy, Precision, Recall, F1-score, and AUC. Results: The VGG16 model achieved the best performance in pneumonia recognition, with an Accuracy of 90.38% and an F1-score of 92.59%. Conclusion: Significant performance differences exist among different CNN models in pneumonia X-ray image recognition tasks, providing a reference for model selection in clinical assisted diagnosis systems.