Research on the Assistive Strategy of Medical Image Segmentation Technology based on Medical Transformer Model in Intelligent Diagnosis of Tuberculosis in Xinjiang Region

Abstract

The aim of this study is to explore the application of Medical Transformer model-based medical image segmentation technology in the intelligent diagnosis of tuberculosis in Xinjiang. Tuberculosis (TB), as a serious infectious disease, is still a major public health problem in Xinjiang and other regions where medical resources are relatively scarce. Traditional tuberculosis diagnosis methods suffer from low efficiency and insufficient accuracy, while deep learning technology provides a new solution for automatic medical image segmentation of tuberculosis. In this paper, a Medical Transformer model based on Gated Axial-Attention is proposed and combined with a local-global training strategy (LoGo) to effectively improve the segmentation accuracy of CT images of pulmonary tuberculosis. The problem of data scarcity in Xinjiang region was alleviated by strategies such as data enhancement, migration learning and cross-region data sharing. The results show that the model not only reduces the computational complexity but also performs well with small datasets, which is suitable for the environment with limited medical conditions in Xinjiang and is expected to improve the efficiency of early screening and diagnosis of TB. In addition, future research can further optimize the performance of the model through model lightweighting and self-supervised learning to enhance its potential application in practical diagnosis