Selection Of Stent Graft And Development Of A Machine Learning Predictive Model Of Bird-Beak Configuration During Zone 2 Thoracic Endovascular Aortic Repair

Part 1 Chimney technique and single-branched stent graft for the left subclavian artery preservation during zone 2 thoracic endovascular aortic repair for acute type B aortic dissectionsBackground Thoracic endovascular aortic repair(TEVAR)has become the first-line therapeutic option for managing complicated type B aortic dissections.However,a healthy proximal landing zone is the premise of the approach and the determinant of its outcomes.Intentional coverage of the left subclavian artery(LSA)to extend the proximal landing zone has been considered to be associated with a high risk of upper extremity ischemia and posterior circulation stroke.Therefore,different techniques and devices,including carotid to subclavian artery bypass,chimney,and fenestrated techniques,have been introduced to manage acute type B aortic dissections with unfavorable proximal landing zones(<15mm or involving the LSA ostium)for the LSA revascularization during TEVAR.Single-branched stent graft for preserving the LSA has been performed to manage aortic arch pathologies.Carotid to subclavian artery bypass carries a high risk of peri-operative mortality and neurologic complications(stroke,spinal cord ischemia,and phrenic nerve palsy).The issue of fenestrated endograft integrity may be related to long-term outcomes.The chimney technique is considered to increase the risk of endoleak.A single-branched stent graft results in reduction of perfusion of the superior aortic arteries with increased risks of occlusion of the branched section.Therefore,the selection criteria for different techniques for preservation of the LSA during zone 2 TEVAR for type B aortic dissections with an unpreferable proximal landing zone has not been established.Objective To evaluate the efficacy and safety of the chimney technique or singlebranched stent graft for the preservation of the LSA during zone 2 TEVAR for acute type B aortic dissections and summarize our single-center experience.Materials and methods From February 2017 to June 2020,137 patients who underwent LSA revascularization during zone 2 TEVAR were enrolled.The chimney technique was indicated for patients with ruptured dissections,and the single-branched stent graft was offered for patients with retrograde intramural hematoma or dissection involving the LSA ostium.Unless otherwise specified,both techniques were offered without preference.The chimney technique was performed in 68 patients(group A),and a single-branched stent graft was deployed in 69 patients(group B).All procedures were performed during the acute phase.Technical success rate,immediate postoperative endoleak,neurologic complications(stroke or spinal ischemia),30-day mortality,12-month clinical success rate,all-cause mortality,patency of the LSA,reintervention,and aortic remodeling were analyzed.Comparing the occurrence of the bird-beak configuration(BBC)and the peri-procedural cost was also performed.Results Primary technique success was achieved in 66 and 67 patients in groups A and B,respectively.The incidence of immediate post-operative endoleak,neurologic complications(stroke or spinal cord ischemia),and 30-day mortality were 5.9%,1.5%,and 4.4%in group A,and 2.9%,2.9%,and 2.9%in group B,respectively.During follow-up,the 12-month clinical success rate was similar(98.4%in group A and 96.9%in group B).All-cause mortality was similar in both groups(3.1%in group A and 4.5%in group B).The patency of the LSA was not significantly different between the two groups,with two occlusions in group A and three in group B.The reintervention rate was higher in group B(3.1%vs.1.6%,P=.544),with a negligible difference.BBC was more prominent in group B,with an incidence of 59.4%,and a significantly higher periprocedural cost was confirmed in group B.Conclusion Both techniques are feasible for LSA preservation during zone 2 TEVAR for acute type B aortic dissections with encouraging outcomes.Despite the higher incidence of BBC for single-branched stent graft,the relatively higher incidence of BBC in group B is discordant with the infrequency of peri-operative and mid-term complications.Served as a minimally invasive alternative,it is necessary to evaluate whether both techniques could offer outcomes comparable to carotid to subclavian artery bypass.Treatment decisions should be made in a multi-professional context,in accordance with characteristics of aortic pathology and patient preference.Part 2 Development of a machine learning predictive model of birdbeak configuration during zone 2 thoracic endovascular aortic repairBackground Thoracic endovascular aortic repair(TEVAR)has been accepted routinely as the first-line therapeutic option for type B aortic dissections,with a lower incidence of morbidity and mortality than open surgery.Although the indication of TEVAR has been extended from descending thoracic aorta to arch pathologies,new emerging complications always require reintervention and can be challenging to predict pre-operatively.The high angulation of the aorta arch makes it challenging to oppose the stent graft to the aortic wall along the lesser inner curve,which results in stent graft deformity defined as bird-beak configuration(BBC).It is responsible for reduction of effective proximal landing zone leading to the increased risk of type Ⅰa endoleak,device collapse and fracture,aortic coarctation,and malperfusion.Both the device’s design and geometric characteristics of the aortic arch affect the underlying mechanics of BBC occurrence,especially for the radial force of the stent graft,inner curvature of the aortic arch,and the morphology of the proximal landing zones.BBC is more prominent during zone 2 TEVAR using single-branched stent graft.Therefore,scrupulous pre-operative planning is required when performing zone 2 TEVAR using single-branched stent graft in patients with a highly angulated aortic arch,and intensive CT scan follow-up should be mandatory to detect the development of BBC.The association between this phenomenon and anatomical and device-related factors remains unclear.A deeper understanding of the morphological factors related to BBC is required to improve the performance and long-term durability of TEVAR.The gold standard method for obtaining geometric features has not been established.Objective The objective of the study was to construct the aortic arch geometric models through thoracic aorta image segmentation and centerline extraction,obtain the geometric feature parameters,and develop a machine learning predictive model of BBC in patients undergoing zone 2 TEVAR using single-branched stent graft.The model might be helpful for risk stratification of BBC and early individual intervention.Materials and methods From October 2019 to March 2022,201 patients with acute type B aortic dissection who underwent left subclavian artery revascularization using single-branched stent grafts during zone 2 TEVAR were enrolled in the study.All pre-and post-operative analyses were performed by two experts with>10 years of experience in TEVAR.According to the presence(Group A)or the absence(Group B)of BBC,these patients were categorized into 2 groups.The data related to patients’demographics,aortic geometric features,and stent grafts were analyzed retrospectively.131 patients,enrolled from October 2019 to February 2021,were used as the derivation cohort,and 70 patients,enrolled from March 2021 to March 2022,were selected for the validation cohort.Firstly,we built a deep learning-based segmentation network U-Net and trained the network with sparse annotations.The trained U-Net was used to segment the region of interest of aortic CT images.Then the centerlines of aortic arches were extracted based on iterative thinning algorithms,and the inner and outer lines of aortic arches were separated based on a gradient projection method.Secondly,the centerline,inner line(inner curve),and outer line(outer curve)were fitted into parametric B-spline curves,respectively,and the key geometric features parameters were calculated.Finally,the relevant parameters of aortic geometric features were selected and analyzed for significance.Based on the commonly used machine learning models,the relevant parameters of 131 cases were used to construct the BBC risk prediction model.We also investigated the feature vector of a patient by continuously sampling multiple curvatures of centerlines.For model validation,70 cases were assigned to the external validation dataset.Results 1)The modified U-Net was trained based on sparse annotations of three classes(regions of interest of aortic arches,other branches,and background).The segmentation results showed that the trained U-Net has learned to segment the region of interest of aortic arches,and the generated aortic arch masks achieved smooth edges.2)The fitted curves of inner and outer lines as well as the centerline of aortic arches were sufficiently smooth.In the experiment,we draw the normal lines of fitted curves with lengths equal to the corresponding curvatures.These lines are consistent with humans’ intuition.3)We built a BBC risk prediction model based on 131 samples with commonly used machine learning methods,mainly including Logistic Regression,Support Vector Machine,Naive Bayes,Decision Tree,and Random Forest.The BBC risk predictive model was trained firstly based on group 1(max and mean curvatures,aortic angulation,tortuosity index,etc.),and the accuracy was only 66.41%,and AUC was 0.6677.To further improve the performance,we constructed a feature vector by sampling 64 curvatures of centerlines(group 2).The trained Random Forest model based on group 2 achieved an average sensitivity of 84.25%,a maximum accuracy of 84.62%,and an average AUC of 0.7303.The average AUC was 0.7 in the external validation dataset.Conclusions Based on CTA,the aortic arch geometric models could be constructed by using trained U-Net to segment the region of interest,and the key geometric feature parameters could be achieved.Continuously sampling 64 points of the centerline curvature was performed,which was fed into the Random Forest model to obtain the final result of BBC risk prediction.Based on the multi-point curvature(64)parameters of the aortic centerline,the Random Forest model obtained high sensitivity and accuracy for pre-operative prediction of BBC occurrence.This model might be helpful for risk stratification of BBC and early individual intervention.For patients with a high risk of BBC occurrence,scrupulous pre-operative planning,more intraoperative caution,and long-term clinical and morphological surveillance are mandatory.