Arterial Input Function Estimation With Blind Source Separation In Prostate DCE-MRI

Prostate cancer ranks first in the global male cancer incidence rate for many years.With the acceleration of population aging and economic development,the number of male prostate cancer patients in China is increasing rapidly.Dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)is a non-invasive imaging detection method that can reflect the physiological information of tissues.Quantitative parameters related to the physiological characteristics of tissues can be obtained by fitting the pharmacokinetic model,which can be used for the auxiliary diagnosis of benign and malignant lesions.However,the actual arterial input function(AIF)as the input signal of the model cannot be actually obtained,the assumption or approximate signals such as the concentration of the iliac artery blood vessel region and the population AIF are generally used instead,and its accuracy affects the results of quantitative analysis.In view of this,the research goal of this essay is to adopt a different idea from traditional DCE-MRI data analysis,proposes a data-driven research idea,and uses a blind source separation algorithm to estimate the AIF in the prostate DCE-MRI data.The main content of this essay is as follows:(1)The non-linear fitting process of the DCE-MRI data conversion formula is simplified into a linear fitting process by using the reparameterization method.Aiming at the effect of image noise on the calculation result of contrast agent concentration,an improved signal intensity contrast agent concentration conversion method is used to convert the signal intensity time curve of the voxel in the prostate area into a contrast agent concentration time curve.Through the reparameterization of the conversion formula,the nonlinear fitting problem of the conversion formula is changed to the linear fitting problem,which reduces the difficulty of data fitting and the influence of abnormal values caused by noise on the concentration conversion.(2)The blind source separation algorithm based on convex geometry analysis is introduced,and the data-driven research idea is adopted to estimate the AIF.Aiming at the dimensionality reduction method based on the minimum error that does not conform to the physical meaning,this essay uses the dimensionality reduction method based on the non-negative matrix factorization under the constraint of the non-negativity of the reconstructed signal.A blind source separation algorithm based on the minimum enclosing simplex volume algorithm is used to estimate AIF.Aiming at the problem that the algorithm is sensitive to noise,this essay adopts two methods,relaxation factor and regular term to make the estimated simplex closer to the simplex composed of the real source signal.Experimental results show that the improved algorithm is more robust to noise.(3)An improved linear mixed pharmacokinetic model is proposed to solve the inapplicability of the linear mixed model used in the above algorithm and the extended Tofts-Kermode model widely used in prostate DCE-MRI.This essay adopts the quantitative analysis of the source signals based on algorithm estimation,avoiding the high calculation amount of the traditional voxel-by-voxel quantitative analysis method.And the feasibility and accuracy of the algorithm are verified indirectly through the difference between the benign and malignant quantification parameters of the constructed simulation data and the actual data.The simulation data results show that the improved algorithm is robust to noise,and the actual data analysis results show the effectiveness of the algorithm in the task of classifying prostate benign and malignant lesions.According to the above results,this essay draws the following conclusions: The blind source component with the characteristics of AIF can be derived from the prostate DCE-MRI data by the blind source separation algorithm based on the data-driven research idea;Using an improved linear mixed pharmacokinetic model,quantitative pharmacokinetic parameters can accurately distinguish benign and malignant prostate lesions.