Research On CTPA Pulmonary Embolism Images Segmentation Based On Multi-view Weighted Fusion Attention Mechanism

Pulmonary embolism is the general term for a group of diseases or clinical syndromes in which various emboli block the pulmonary artery or its branches.Its harmfulness is second only to myocardial infarction and stroke.The clinical symptoms and signs of pulmonary embolism are usually not specific,which poses greater challenges and higher requirements for clinicians’ timely and correct diagnosis.CT pulmonary angiography(CTPA)has high sensitivity and specificity for the diagnosis of PE,and is non-invasive and convenient.It has become the first choice for the diagnosis of PE.However,the examination of PE requires radiologists to carefully track whether there is suspicious PE in each pulmonary artery in a large number of CT images.High-intensity work will greatly increase the probability of misjudgment and missed judgment.With the updating and iteration of computer software and hardware,the theory and practice of pattern recognition and medical image processing continue to accumulate.Through powerful deep learning algorithms,it is possible to establish accurate medical auxiliary judgment models to assist doctors in diagnosis and treatment.Based on the above facts,based on CTPA image data,this study carried out a targeted study on the segmentation algorithm of pulmonary embolism.The main work of this paper is as follows:(1)Based on the most commonly used segmentation network U-Net and its variant network in the current medical image segmentation field,a segmentation method suitable for pulmonary embolism images is designed,the performance of the classic U-Net and its variants in segmenting pulmonary embolism images are analyzed,and the existing shortcomings and limitations of the network are discussed.(2)Aiming at the problem that the two-dimensional segmentation network is prone to misjudge the suspected pulmonary embolism area and the three-dimensional segmentation network consumes huge computing resources,a network structure based on the multi-view weighted attention mechanism is proposed(MWA U-Net).This model uses three parallel feature extraction networks to perform feature extraction on three views respectively.By introducing an attention mechanism and using adaptive weighting of multi-view collaboration to imitate clinical observation and diagnosis,it can effectively improve the accuracy of segmenting pulmonary embolism lesions.