A pneumonia detection method based on the variant structure of broad learning system

Pneumonia as a common respiratory disease, its accurate and rapid diagnosis is crucial to the health of patients. With the innovation of medical technology and the development of artificial intelligence, computer-aided diagnosis has been increasingly used in the medical field. Deep learning has achieved remarkable results in the field of pneumonia detection, but its large number of parameters and complex network structure lead to limitations such as long training time and high consumption of computational resources. To solve the above problems, a pneumonia detection method based on the variant structure of broad learning system is proposed in this paper. The method introduces the cascade pyramid structure on the basis of original broad learning system. Meanwhile, the pre-trained EfficientNet network is utilised as the front feature extractor. In addition, the incremental learning algorithms applicable to the model are proposed in this paper, including adding additional enhancement nodes, feature nodes and training samples to further optimise the model performance. Finally, comparative experiments are conducted on the publicly available dataset of chest X-rays for pneumonia. The experimental results show that the method in this paper achieves 92.83% accuracy and 98.86% AUC value, which are comparable to many deep convolutional neural networks, while the training time of the model is significantly shortened.