基于混合深度学习框架的气胸图像分类算法

Pneumothorax Image Classification Algorithm Based on Hybrid Deep Learning Framework

为了更好地解决卷积神经网络(CNN)其有限的局部感受野,限制其在气胸图像分类中的性能等问题,提出一种基于CNN和Vision Transformer的气胸分类模型。首先,依托三个开源胸部数据集完成实验数据的清洗与收集,然后,将VIT-L/16的MLP模块进行改进,最后,利用改进的VIT模型与CNN模型进行加权分类得到最终的结果。实验结果表明该模型在二分类任务中达到了99%的准确率,以及0.99的AUC分数,与其它模型相比,具有更好的分类性能。

In order to better solve the problems of the limited local receptive field of convolutional neural network(CNN)and its performance in the classification of pneumothorax images,a pneumothorax classification model based on CNN and Vision Transformer is proposed. Firstly,the cleaning and collection of experimental data is completed by relying on three open-source chest datasets. Then,the MLP module of VIT-L/16 is improved. Finally,the improved VIT model and CNN model are used for weighted classification to obtain the final result. The experimental results show that the model achieves an accuracy of 99% in the binary classification task and an AUC score of 0.99,which has better classification performance than other models.