Computer Aided Diagnosis Based On Texture Features Of Liver MRI

Liver cirrhosis and fibrosis are two of the most common chronic liver diseases in clinical practice.Liver fibrosis is a chronic hepatic lesion,which is characterized by a multi-reason caused abnormal proliferation of connective tissue in the liver.Cirrhosis is a diffuse liver damage which generally due to one or more repeated or long-term causes.Imaging examination has become a mainstream way of checking liver cirrhosis and fibrosis.Especially,because of the advantages such as no ionizing radiation,multi-parameter,multi-orientation,multi sequence and high-resolution,MRI has become one of major approaches to diagnose liver disease.In general,texture features of MRI is analyzed by radiologists to determine health status of liver.This diagnostic method depends on doctors' experience and priori knowledge,so clinical diagnosing is lack of objective basis.Therefore,in this thesis,a computer-aided diagnosis system was proposed to stage Liver cirrhosis and liver fibrosis based on MRI images by analyzing the different texture feature of liver.The main work of this thesis is as follow:(1)Human liver cirrhosis staging based on MR8 texture feature clustering analysis.The surface of normal liver is smooth and has no granular texture in MRI images.However,cirrhosis appears crack,scattered and high signal around the nodules.Texture features can be fully extracted by MR8 filters.So,in this thesis,MR8 filters was used to staging human liver cirrhosis with clustering analysis due to the advantages of rotation invariance and lower dimension Five sequences of liver MRI images was used in cirrhosis staging.The results of ROI staging showed that the accuracy of T2 sequence and equilibrium phase were 100%.The accuracy of T1 sequence,arterial phase and portal venous phase also reached 95%.The staging results of all cases using voting mechanism were 100%.This method took fully advantages of MRI images and was also easy to understand and operation.In addition,the computation complexity of our method is small,while the accuracy of liver cirrhosis staging is high.Therefore,this method we presented can be an assistant way of diagnosis in clinical practice.(2)Rat hepatic fibrosis computer-aided diagnosis(CAD)based on T1 sequence of MRI.A classical method combined Gray-level Co-occurrence Matrix(GLCM)with Back Propagation Neural Network(BPNN)was used to stage liver fibrosis of rats.However,the performance was not good enough.To solve this problem,in this thesis we presented a novel method based on Clusterdp clustering to stage hepatic fibrosis of rats with T1 sequence of MRI.The results showed that the Clusterdp method had better performance on staging hepatic fibrosis of rat compared with the classical method.Especially in S0,S2 and S4 stages,the staging results of ROI reached 100%.But the results of S1 and S3 stages were not satisfactory.The results of S1 and S3 stage were 89.66% and 83%,respectively.Combined with several ROIs of each rat,the final staging results of all stage reached 100% except for S3.Only one subject in S3 was wrongly staged.In contrast,the performance of BPNN based on GLCM was far less effective than the method we proposed.Based on the above results,the method we presented in this thesis showed good performance on staging hepatic fibrosis of rats.Therefore,this novel method based on Clusterdp have the clinical significance for staging hepatic fibrosis.