Research On Liver Tumor Segmentation Algorithm Based On Deep Learning

As one of the indispensable organs of the human body,the liver undertakes the function of metabolizing various substances in the human body,and is also an important heat supply organ.With the fast pace of life in today’s society and people’s abnormal living habits,the number of liver cancer patients is increasing.CT image is based on the different absorption coefficients of X-rays by different organs or tissues.Compared with other technologies,CT imaging technology is relatively mature and has high sensitivity.It can more accurately present the detailed characteristics of the body’s organs,tissues and lesions,enabling doctors to observe more clearly and formulate effective treatment plans in time.Therefore,it is commonly used.in the diagnosis of liver tumors.Accurate segmentation of liver and liver tumors based on medical imaging is an important basis for doctors’ diagnosis and surgical treatment.Especially in smart medicine,accurate segmentation of liver and liver tumors is the premise of other treatment steps,and it is also an important basis for doctors to diagnose the disease and determine the treatment plan.Traditional segmentation methods require doctors to manually label the lesion location with corresponding software based on pathological knowledge and relevant clinical experience,which is very timeconsuming and labor-intensive.In addition,the accuracy of labeling is determined by the doctor’s subjective judgment and professional background knowledge,and there are great differences and instability.Therefore,more accurate automatic segmentation tools are needed,and computer-assisted segmentation can be achieved more quickly and accurately.In recent years,with the rapid development of machine learning technology,more and more scholars have introduced deep learning technology into the field of medical image processing,which has improved the accuracy of medical image segmentation.The tumor is so small that it is difficult to distinguish with the naked eye,and the location of the tumor is also variable,so it is still difficult to accurately segment the tumor.In view of the above problems,the following research work is carried out in this paper:1.An image segmentation network model based on multi-scale feature extraction is proposed for liver segmentation.The model uses a redesigned multi-scale feature extraction structure,which includes multiple residual connections and convolution blocks with different receptive fields.Residual connections can alleviate the problem of network overfitting,while continuous convolution operations will lead to semantic the loss of information and the addition of residual connections can also suppress the loss of semantic information.The size of the liver in the CT slice is different and the position is changeable.It is difficult to extract comprehensive features with the fixed receptive field of traditional convolution.Therefore,different sizes of receptive fields are used.Multiple convolutional blocks to extract more comprehensive features.The experimental results show that the model has better segmentation effect than other models.2.For the more difficult liver tumor segmentation,this paper proposes an image segmentation network model that integrates multiple attention mechanisms.In the encoding stage,the model first uses depth wise separable residual convolution to extract features.The depth wise separable convolution has less parameters and thus can speed up the training process of the network,and then uses an attention module that integrates multiple attention mechanisms to adjust the features.The weight of the value,so that the network model pays more attention to the area where the liver tumor is located,and reduces the influence of background and irrelevant semantic information.Experiments show that the performance of this model for segmenting liver tumor is better than that of the compared segmentation models.