The Clinical Application Research Of Computational Fluid Dynamics And Machine Learning In Evaluation Of Coarctation Of Aorta And Aortic Stenosis Patients

There are high morbidity,high disability and high mortality in cardiovascular disease.Coarctation of aorta?Co A?and aortic stenosis?AS?are common cardiovascular diseases,there is no effective medication for CoA and AS for now.It is known that diagnosis and evaluation of Co A and AS have many limitations.Hemodynamics are clinical reference for diagnosis and evaluation of Co A and AS,and invasive cardiac catheterization is gold standard for acquiring hemodynamics.Up to now,there are some limitations in noninvasive method for obtaining hemodynamics:transthoracic echocardiography?TTE?is known to overestimate the pressure gradient although it is an accurate velocity measurement;MRI is expansive and has a complex examining progress.Computational fluid dynamics?CFD?is a research focus in cardiovascular system,it is noninvasive,has high resolution and much more hemodynamic parameters.In recent years,machine learning was also one of the research focuses in cardiovascular system.It is noninvasive,moreover,it has high operation speed and strong reusable property.MDCTA is wildly used to evaluate the anatomy,but it is unable to perform hemodynamic analysis.The main task and results are list as follow:1.Validate the feasibility of CFD method in human aortic hemodynamic assessment.This study included 25 patients with normal aorta.All patients underwent both transthoracic echocardiography?TTE?and MDCTA within two weeks prior to cardiac catheterization.CFD models were created from MDCTA raw data.Boundary conditions were confirmed by peak systolic velocity?PSV?at aortic valve from TTE and pressure from limb measurement.Lumped parameter model?LPM?was used to calculate the pressure if the pressure was unknown.Peak systolic pressure?PSP?derived from CFD(PSP_CFD)was compared to catheterization(PSP_CC),while the PSV derived from CFD models(PSV_CFD)was compared to TTE measurements(PSV_TTE).According to research results,PSP_CFD and PSV_CFD showed good agreements between PSP_CC?r=0.918,p<0.001;mean bias=-1.400 mmHg?and PSV_TTE?r=0.968,p<0.001;mean bias=-7.68 cm/s?.In addition,the distribution of hemodynamics can be visualized by CFD.Some immeasurable parameters like wall shear stress and helical flow can be also visualized.This study illustrated the feasibility and correctness of CFD in aortic hemodynamic assessment.2.Use the MDCTA-based CFD to validate the noninvasive diagnosis of Co A.The study considered 75 pediatric patients divided into a training set?n=50;25 with Co A and 25without Co A?and a separate testing set?n=25?.CFD models were created from CTA raw data.Boundary conditions were confirmed by PSV at aortic valve from TTE and pressure from limb measurement.LPM was used to calculate the pressure if the pressure was unknown.The diagnostic performance of CFD was evaluated by ROC curve.According to research results,CFD can distinguish normal aorta between aorta with CoA.In training set,peak systolic pressure gradient?PSPG?,PSV and average peak systolic wall shear stress?AMWSS?had high diagnostic efficiency?AUC=0.987,0.931,0.978?,the sensitivity were 92%,92%,88%respectively and the specificity were 92%,96%,80%respectively.The false-negative results were only 2,3,2 respectively.In testing set,PSPG,PSV and AMWSS derived from CFD had relatively high AUCs?0.953,0.947,0.823,respectively?.The sensitivity were 93%,87%,80%respectively and specificity were 90%,80%,100%respectively.The negative-false rates were 1/15,3/15,2/15.This study demonstrated that the noninvasive MDCTA-based CFD method can accurately diagnose Co A.3.Use MDCTA-based CFD method to implement the grading of AS.This study employed 30 patients?10 mild stenosis patients,10 moderate stenosis patients and 10 severe stenosis patients?.All patients underwent both TTE and MDCTA within two weeks prior to cardiac catheterization.CFD models were created from MDCTA raw data.Boundary conditions were confirmed by PSV at left ventricular outflow tract from TTE and pressure from limb measurement.LPM was used to calculate the pressure if the pressure was unknown.The grading efficiency was evaluated by multinomial logistic regression and weighted Kappa test.According to results,CFD method had relatively high accuracy?86.67%?,and the weighted Kappa test indicated that there was exactly high consistency of grading between CFD and catheterization??=0.80,p<0.001?.This study demonstrated that CFD method can grade the Severity of AS.It is the expansion of CFD application scenario.4.Use the machine learning algorithm to validate the its diagnostic performance in Co A.This study included 66 patients?medium age:12 months[range:1 month-17 years],54.5%male?.All patients underwent both TTE and MDCTA within two weeks prior to cardiac catheterization.K-Nearest Neighbor?KNN?,Naive Bayes?NB?and Random Forest?RF?algorithm were employed in this study,the features used in learning were derived from TTE and MDCTA,and the labels were confirmed by catheterization.Features were used to train the algorithm after pre-processing.Ten-fold cross validation was used to validate the algorithm.ROC and F-measure were utilized to evaluate the predict effect.According to the results,there were good accuracy?0.812,0.855,0.913 respectively?using KNN,NB and RF.The AUCs were 0.900,0.936,0.965 respectively,and the F-measure were 0.814,0.854,0.897respectively.This study indicated that machine learning algorithm can diagnose Co A accurately.