To Evaluate The Predictive Value Of CTA-based Morphological Parameters And Radiomics For High-risk Infrarenal Abdominal Aortic Aneurysm

Purpose:The progression and rupture risk evaluation of abdominal aortic aneurysm(AAA)is largely dependent on the maximum diameter,whereas part of the small aneurysm has the possibility of rupture,with rupture rates as high as 23%for aneurysms less than 5.0 cm in diameter,while some larger aneurysms can remain stable overtime.This reflects the significant limitations of diameter-based treatment protocols,and such management measures may lead to late intervention in small aneurysms or unnecessary intervention in stable aneurysms.This study aims to identify potential AAA events requiring surgical repair or rupture by measuring the rate of area change of AAAs and by extracting aneurysm radiomics features.Methods:A retrospective collection of 176 AAA patients from six centers,including Shandong Provincial Hospital,was conducted between July 2011 and June 2021.Patients underwent at least two preoperative aortic computed tomography angiography(CTA)examinations,with two aortic CTA examinations at least 3 months apart.The endpoint event was defined as AAA rupture or/and repair.Patients were divided into event and no-event groups based on at least one year of follow-up.All images were transferred to Siemens semi-automatic image segmentation software for processing,and the maximum AAA aneurysm area and maximum lumen area were measured from baseline and follow-up aortic CTA images,respectively,and the aneurysm area change rate(AACR)and lumen area change rate(LACR)were calculated based on the area change before and after and the interval.The area of interest of the baseline images was outlined using ITK-SNAP 3.8.0 software.Three types of areas of interest were outlined,including the lumen of the aneurysm,the intraluminal thrombus(ILT),and the entire aneurysm.All segmented regions of interest were automatically subjected to image histological feature extraction using the toolkit PyRadiomics.The least absolute shrinkage and selection operator(LASSO)-Cox was used for optimal feature selection in generating the radiomics score(Radscore).A Cox proportional risk model was used to predict the event risk in patients with infrarenal AAA.Six Cox models were built from the training set for analysis,including the ILT-based Radscore model,the lumen-based Radscore model,the aneurysm-based Radscore model,the AACR model,the LACR model,and the merged model(a combination model of the above five models).Two additional adjusted Cox models were developed to determine the correlation between predictors(radiomics predictors,AACR and LACR)and events in patients with infrarenal AAA.The first model had adjusted factors that included age and gender.The second model added additional adjustment factors for smoking,height,weight,body mass index(BMI),diabetes,hypertension and hyperlipidemia to the first model.The C-statistics were utilized to evaluate the predictive performance of the Cox model,Kaplan-Meier analysis was conducted to stratify the risk groups,and calibration curves were used to test the accordance between the actual and predictive five-year event risks for the different predictors.Results:The study ultimately included 130 patients with infrarenal AAA,116(89.2%)males and 14(11.8%)females,with a median age of 70(64,76)years.The training set included 76 patients and the external validation set 54 patients.AAA-related events occurred in 55 cases with a mean follow-up time of 1266 days(minimum 366 days,maximum 3505 days)and an interval of 565 days(minimum 97 days,maximum 2650 days)for CT scans.On the external validation set,the C-index for ILT,lumen,aneurysm,AACR,LACR,and merged models were 0.719(0.610-0.828),0.642(0.506-0.778),0.679(0.550-0.808),0.614(0.482-0.747),0.607(0.466-0.749)and 0.752(0.653-0.851).Kaplan-Meier analysis showed that in the external validation set,the lumen model was not statistically significant for dividing patients into two groups(Log-rank p=0.068),while the aneurysm(Log-rank p=0.024),AACR(Log-rank p=0.013)and LACR(Log-rank p=0.043)models showed better stratification performance.ILT(Log-rank p=0.0047)and merged(Log-rank p=0.0025)models showed the best stratification performance.The results of the adjusted Cox model showed that the predictors in this study(radiomics predictors,AACR and LACR)were all independent risk factors for events in patients with infrarenal AAA(p<0.05).There was a significant difference in AARC and LARC between the non-event and event groups(AARC:1.2 cm2/y vs.2.5 cm2/y,p=0.008;LARC:0.2 cm2/y vs.0.5 cm2/y,p=0.03).Conclusions:The radiomics model of ILT and the merged model had the best predictive performance for AAA events,with AACR and LACR significantly higher in the event group than in the non-event group.CTA-based radiomics,AACR and LACR have good predictive value for high-risk events in patients with infrarenal AAA,which is important for developing clinical management strategies.