Research On Segmentation Of Colon And Aided Diagnosis Of Aortic Dissection In Ct Images Based On Deep Learning

In the medical field,the tension between the massive of medical image examinations and the scarcity of professional doctors has become increasingly prominent.Therefore,how to analyze medical images rapidly and effectively has become a vital issue.Artificial intelligence is a breakthrough point of problems mentioned above by virtue of its highefficiency and powerful capacity,which can improve the efficiency and accuracy in medical field.In this paper,we focus on two academic problems with research value and unique characteristics,based on deep learning algorithms.The intelligent segmentation of colon is an unstable organ segmentation problem,placing emphasis on the targeted optimization and resolution of this difficult problem.The aided diagnosis of aortic dissection is a fatal and rare disease diagnosis problem,placing emphasis on the comprehensive analysis and resolution of this complex problem.In gastroenterology,colon segmentation is a key task routinely,of which accuracy directly influences the quality of the radiotherapy plan.Due to the fact that the position and shape of colon vary among people,the effect of intelligent colon segmentation needs improving.In this paper,we propose a specific model,called SAN U-Net,to improve the accuracy of targeted colon segmentation in CT images.Firstly,SE-Attention module is designed to enhance effective information in the aspect of channel and scale.Secondly,making advantages of multiple-receptive-fields feature fusion,the semantic gap can be filled.Finally,Non-Local module is employed to excavate deep features to assist better segmentation.The experimental results reveal that SAN U-Net model is more suitable for colon intelligent segmentation than other existing models,increasing the segmentation effect of colon in CT images effectively.In cardiovascular surgery,aortic dissection seriously brings a high risk to patients with its rapid onset and fatal severity.As a relatively rare disease,aortic dissection faces insufficient assisted diagnosis research.In this paper,we put forward an end-to-end automatic diagnosis scheme of aortic dissection,aiming to meet the actual medical demands from multiple perspectives.Firstly,Mask R-CNN is utilized to multi-object segmentation of aorta,followed by the post-processing module with medical knowledges.Secondly,in view of the spatial continuity among CT images,a screening algorithm is proposed for effect promotion.Finally,we achieve multi-level diagnosis by means of ResNets.Compared with other methods,we consider the possibility of aortic arch lesions.The experiments show that our scheme performs decently in slice-level and patient-level assisted diagnosis of aortic dissection,both reaching more than 96%.