Attention Mechanism-based Identification Of Liver Cirrhosis

Liver cirrhosis has become one of the common chronic diseases,threatening people's life and health.Therefore,the research on the identification of cirrhosis has more practical significance and application value.Its clinical diagnosis mainly adopts ultrasound imaging technology to scan and present the pathological areas of the liver,and then doctors make subjective diagnosis of the patient's liver ultrasound images.The resulting diagnostic results are often subjective and low reliability.With the rapid development of digital medical equipment,computer image processing and deep learning methods,computer-aided diagnosis technology for medical image analysis has brought new research directions for the diagnosis of liver diseases,and provided more opportunities and challenges for the identification of cirrhosis.At present,deep learning algorithm is the mainstream method of computer-aided diagnosis for the recognition of cirrhosis.Therefore,the deep learning algorithm and the main theory of attention mechanism were adopted in this paper to enhance the ability of the network model to learn the characteristic information of liver images,thus improving the recognition accuracy of cirrhosis disease.This paper mainly proposes three methods for the identification of cirrhosis:(1)A cirrhosis identification method using optimized Alexnet model and extrusion excitation network SENet was proposed.First,in view of the particularity of the liver image size,adjust Alexnet network parameters,and then add in the Alexnet SENet module,the module is to strengthen the effect of convolution weights of neural network to channel information,extract more has the characteristics of directional information,more conducive to get cirrhosis of the liver in the image texture information,by the experimental results show that this method improved the CNN recognition rate,but also found that the addition of the module adds the parameters of the network.(2)To solve the problems found in the above experiments,this paper adopts a more lightweight attention module ECA.ECA module only adds a small number of parameters,and selects the size of one-dimensional convolution kernel to strengthen cross-channel information interaction,which significantly reduces the complexity of the model and maintains the performance of the network.Secondly,by combining the module with Mobile Net V2,the recognition rate of the method is improved by 4%through experimental comparison,and the test time is significantly reduced.(3)On the basis of the above two methods,this paper further improved the model by using a spatial channel squeeze excitation model scSENet for cirrhosis identification.This model is a variant of SENet,which is composed of SSE and CSE,and also takes into account the spatial and channel relationships of the network.First of all,based on MVGG(the improved VGG16)and then integrated with scSENet module,this paper not only optimized the complexity and operating efficiency of the network,but also further improved the recognition effect of cirrhosis images.The experiment showed that the best recognition rate reached 98.78%.The combination of attention mechanism and deep learning algorithm is more conducive to the recognition of liver lesion areas and the research and judgment of attending physicians,which has higher clinical value.