TransCotANet: A Lung Field Image Segmentation Network with Multidimensional Global Feature Dynamic Aggregation

Chest X-ray (CXR) images can be used to diagnose a variety of lung diseases, such as tuberculosis, pneumonia, and cancer. However, the variation in morphology due differences age, gender, severity pathology makes high-precision segmentation challenging task. Traditional networks, U-Net, have become standard architecture achieved remarkable results field image tasks. because traditional convolutional operations only explicitly capture local semantic information, it is difficult obtain global resulting performance terms accuracy requirements medical practical applications. In recent years, introduction Transformer technology natural language processing has great success computer vision. this paper, new network called TransCotANet proposed. The based on U-Net with neural networks (CNNs) backbone extracts information through symmetric cross-layer connections encoder structure, where stage includes an upsampling module improve resolution feature map, uses dynamic aggregation CotA dynamically aggregate multi-scale maps finally more accurate results. experimental show that method outperformed other methods for datasets.