A Hybrid Lightweight Network based on State Space Model for Rapid Diagnosis of Emerging Infectious Diseases.

During outbreaks of emerging infectious diseases (EIDs), short-term surges in patient numbers often lead to medical resources panic squeeze, hindering timely diagnosis and patient isolation, thereby severely impacting patient's prognosis and impeding disease spread. Traditional clinical diagnosis of EIDs primarily depends on physicians' expertise, which may result in discomfort and healthcare-associated infections, inadequately addressing diagnostic demands during sudden epidemic surges. Meanwhile, most existing studies in EID automatic diagnosis are focused on improving the accuracy without considering resource limitations. In order to meet the surge in demand for short-term diagnosis during the epidemic, we introduced a lightweight network, which uses a structure similar to the MobileViT, alternating the MV2 module and the designed MobileMamba block. In the MobileMamba block, convolution is used to extract local features, while the group SSM module is used to obtain global features with fewer parameters. The proposed method has achieved state-of-the-art performance on the two EID datasets (including COVID-19 and monkeypox datasets) with a very small number of parameters (about 1/73 of ResNet 50, 1/3 of MobileViT, and 1/69 of VMamba) and calculations (about 1/30 of ResNet 50, 1/2 of MobileViT, 1/5 of VMamba), outperforming existing diagnostic models across key metrics such as accuracy, F1 score, and AUC.Clinical relevanceThis study proposes a high-performance and lightweight diagnostic model, which greatly promotes the rapid diagnosis of EIDs.