Detection of Pulmonary Embolism Based on Receptive Field Amplification and Attention Mechanism

Abstract

Pulmonary Embolism (PE) is a serious threat to human life and health due to its high incidence rate and mortality. It is important to detect PE in time for the treatment of the disease and recovery of patients. Computed Tomography (CT) images are often used for disease diagnosis in clinical practice. Existing lung CT image disease classification algorithms only focus on local information, resulting in low accuracy, To solve this problem, a network model (DR-SENet) based on receptive field amplification and attention mechanism is proposed to detect PE. Specifically, Resnet network is used as the backbone network to slow down gradient explosion and gradient disappearance. Channel attention mechanism is used to extract the weight information between feature channels to guide the network to focus on important feature information, At the same time, the receptive field amplification structure is introduced to enhance the feature extraction ability of the network, obtain more comprehensive features, and make up for the limitation of convolution operation focusing on local features. The method is tested on the open PE dataset–FUMPE. Through experiments, we found that our method obtained better index and improved the auxiliary diagnosis performance of pulmonary embolism.