Research on Attention Mechanism Based Assisted Diagnosis of Pulmonary Embolism

The traditional diagnosis of pulmonary embolism (PE) requires doctors to distinguish between computed tomography (CT) images, which is a very time-consuming task and may result in patients not receiving timely and effective treatment. Therefore, the use of computer intelligence to assist in diagnosis is crucial. In this paper, we propose a Coordination and spatial attention convNext (CSACNet) based on CSA attention mechanism to realize early intelligent auxiliary diagnosis of pulmonary embolism disease. First, Convolutional neural network uses ConvNext network as the backbone network for feature extraction, which can solve the gradient problem caused by network deepening. Secondly, the CSA attention module introduced is a fusion of CoordAttention and Spatial Attention, extracting channel, position, and spatial information between features to obtain more discriminative features and improve the network’s classification accuracy for pulmonary embolism images. The proposed method was tested on the largest publicly contested PE dataset(RSNA-STR), and the experimental results showed that it outperformed the current best method and improved the performance of pulmonary embolism assisted diagnosis.