Research On Computer-aided Diagnosis Of Liver Fibrosis And Early Cirrhosis

Liver fibrosis is a kind of pathological process of abnormal connective tissue hyperplasia in the liver,which is usually caused by various pathogenic factors.Any liver's injury will occur fibrosis in the process of liver repair and healing.However,if the injury can not be removed in time,the development of liver fibrosis to the later serious stage will lead to liver cirrhosis with diffuse damage,and then lead to serious complications such as liver cancer.The diagnosis of liver fibrosis and cirrhosis is usually divided into serum examination,liver biopsy and imageology examination.Liver biopsy is invasive.CT has insufficient spatial resolution and radiation damage.MRI is unable to carry out real-time dynamic examination and requires high cost.Ultrasound imageology technology has the advantages of nondestructive,real-time,convenient and nonionizing radiation damages,which is suitable for large-scale screening and follow-up,and has obvious advantages in the diagnosis of liver fibrosis and cirrhosis in comparison.In this paper,computer-aided diagnosis methods of liver fibrosis and cirrhosis based on the high-frequency ultrasound images is proposed.In this paper,the enhancement of medical image data,the segmentation of liver cirrhosis lesion and the four classification problems of liver fibrosis are discussed,which provide an effective method for clinical computer-aided diagnosis of liver fibrosis and liver cirrhosis.The work and innovations of this paper are as follows:(1)According to the characteristics of ultrasound medical image,such as fuzzy and noise pollution,a Residual Dense Conditional Generative Adversarial Network(RDC-GAN)is proposed to enhance the data of medical image,and the goal of obtaining high-quality superresolution image after the medical image is magnified 4 times is realized.The algorithm is evaluated from the aspects of peak signal-to-noise ratio,structural similarity,MOS score and practical application.In the task of liver cirrhosis' s diagnosis,the accuracy rate of using the improved quality data to diagnose mild and severe liver cirrhosis is 5.50% and 7.70% higher than that of the previous data set,and the F1-score of mild,moderate and severe liver cirrhosis is 5.71%,2.54% and 3.99% higher than that of the previous data set,which verifies the practicability of the algorithm.(2)According to the diffuse characteristics of liver cirrhosis,an improved semantic segmentation algorithm based on Normalized cut is proposed.Using the known patches' classification and confidence information to match the original image,mark and correct the isolated points of the original image's lesion area,and add space information constraints with practical physical significance to the minimum energy function of the segmentation standard to realize the segmentation of the liver cirrhosis lesion area,which provides the doctors with the information of clinical diagnosis value.(3)According to the classification of liver fibrosis diseases,Res Net-V2 network based on transfer learning strategy is proposed for the four classification of liver fibrosis.For the pre-trained network model,the weight parameters of each layer are initialized randomly,and the dataset expanded by the method of rotating and cutting patch is added to fine tune various parameters.After softmax classifier and patches' voting principle,the sensitivity of staging of liver fibrosis S0 ? S1,S2,S3 and S4 is 93.75%,96.88%,87.50% and 91.30% respectively;the specificity is 93.75%,93.94%,95.45% and 87.50% respectively;F1-score is 93.75%,95.39%,91.30% and 89.36%,respectively.In the task of quantitative diagnosis of liver fibrosis based on high-frequency ultrasound images,it has achieved good results.Through comparative analysis,this method is superior to other existing methods,and provides a feasible method for clinical computer-aided diagnosis of liver fibrosis.