Prognosis Analysis Of Endovascular Treatment And Research On Intelligent Segmentation And Reconstruction Methods For Uncomplicated Type B Aortic Dissection

Part I Analysis of risk factors for poor prognosis of uncomplicated TBAD treated with endovascular treatmentObjective: The optimal treatment strategy for uncomplicated Stanford type B aortic dissection(TBAD)is still controversial.Stent-related adverse events have prevented thoracic endovascular aortic repair(TEVAR)from becoming the first-line treatment modality for uncomplicated TBAD.The aim of this study was to explore the risk factors for endoleak,retrograde type A aortic dissection(RTAD),and distal stent-induced new entry(d SINE)after TEVAR for uncomplicated TBAD in our hospital over the past decade by analyzing their clinical and imaging data.It is hoped that the treatment effect of TEVAR can be maximized.Methods: The clinical and imaging data of uncomplicated TBAD patients who received TEVAR treatment in our hospital from January 2010 to December 2018 were collected retrospectively.The inherent morphological characteristics of the aorta(aortic arch type,aortic arch radius of curvature,thoracic aortic tortuosity index,taper ratio,etc.),anatomical features(entry tear size,extent of involvement,distance from the primary entry to the left subclavian artery(LSA),thrombosis of the false lumen,etc.)and TEVAR-related local features(stent oversizing,stent length,proximal diameter of the stent,the position of the proximal landing zone,the angle between the distal end of the stent and the continuous artery,the mismatch ratio of the distal end of the stent,etc.).The occurrence of postoperative stent-related adverse events was evaluated by reviewing CTA images.Competing risk analysis was used to assess procedure-related adverse events and estimate cumulative reintervention rates using all-cause mortality as a competing risk.The risk factors for adverse events were determined by univariate and multivariable Cox proportional hazards regression models.Results: A total of 226 patients with uncomplicated TBAD treated with TEVAR were included in this study,with a mean age of 53.1 ± 10.7 years;83.6% were male.The median follow-up time was 4.6 years(interquartile range: 3.8-6.5 years).A total of 24(10.6%)stent-related major complications were observed during follow-up: 24(10.6%)type I endoleaks,7(3.2%)RTAD and 16(7.1%)d SINE.The proportion of men in the type I endoleak group was significantly less than that in the group without type I endoleak(P=0.008).The radius of curvature of the aortic arch in the type I endoleak group was significantly smaller than that in the group without type I endoleak(P=0.003).The distance between LSA and the primary entry tear in the type I endoleak group was 15.4 ±9.4 mm,which was significantly lower than 21.9 ± 13.8 mm in the group without type I endoleak(P=0.018).Multivariable Cox regression models showed that female(Hazard Ratio [HR],2.66;95% Confidence Interval [CI],1.12-6.34,P=0.031),decreased radius of curvature(HR,0.91;95% CI,0.83-0.99,P=0.020)and decreased distance between LSA and the primary entry tear(HR,0.95;95% CI,0.91-0.99,P=0.027)were independent risk factors for type I endoleaks after TEVAR in patients with uncomplicated TBAD.All 7patients in the RTAD group were male and all had acute TEVAR.Proximal oversizing was significantly greater in the RTAD group than in the group without RTAD(P=0.006),at 9.2± 5.4 % and 4.4 ± 4.3 %,respectively.Multivariable Cox regression models revealed excessive proximal oversizing(HR,1.16;95% CI,1.04-1.30,P=0.006)as an independent risk factor for RTAD after TEVAR in patients with uncomplicated TBAD.A median time to d SINE of 24.0 months was observed in 16 cases on follow-up CTA images.Multivariable Cox regression analysis revealed type III aortic arch(HR,7.27;95% CI,1.30-40.81,P=0.024),increased distal oversizing(HR,1.04;95% CI,1.00-1.07,P=0.039),decreased angle(HR,0.97;95% CI,0.96-0.99,P=0.006)and increased distal mismatch ratio(HR,4.19;95% CI,1.33-13.22,P=0.015)were independent risk factors for d SINE after TEVAR in patients with uncomplicated TBAD.Conclusions: Female,decreased radius of curvature and distance between LSA and primary entry are independent risk factors for type I endoleak after TEVAR for uncomplicated TBAD.Excessive proximal oversizing is an independent risk factor for RTAD after TEVAR and acute phase may influence the development of RTAD.A type III aortic arch,excessive distal oversizing,decreased angle between distal stent and continuation artery and excessive distal mismatch ratio are independent risk factors for d SINE after TEVAR in patients with uncomplicated TBAD.Part II Impact of timing and stent oversizing of endovascular treatment on long-term prognosis in uncomplicated TBADObjective: The optimal timing of endovascular treatment for uncomplicated TBAD and the choice of stent oversizing are two key issues to improve the treatment effect of TEVAR.The purpose of this study is to analyze the clinical and imaging data of patients with uncomplicated TBAD in our hospital over the past decade,and to explore the effect of the time from the onset of dissection to TEVAR and different stent oversizing on the early and long-term prognosis after TEVAR in patients with uncomplicated TBAD.Methods: Clinical and imaging data of patients with uncomplicated TBAD treated with TEVAR at our hospital from March 2008 to December 2018 were retrospectively collected.Treatment timing subgroups were divided into: acute TEVAR group(≤14 days)and subacute TEVAR group(15-90 days)according to the time from the first symptom to the TEVAR treatment.The stent oversizing subgroup was divided into: ≤5% oversizing group and >5% oversizing group according to proximal oversizing.Early outcomes focused on mortality and adverse event rates in the hospital and at 30 days after TEVAR.Late outcomes included primary endpoints(all-cause death and aortic-related death)and secondary endpoints(procedure-related complications or a composite endpoint of combined all-cause death,including retrograde type A dissection,endoleak,stent-distal induced new entry tear,and late reintervention).Inverse probability treatment weighting(IPTW)was used to balance baseline variables.Landmark analysis was used to assess the time dependence of treatment effects.Competing risk models were used to assess the incidence of procedure-related complications using all-cause death as a competing risk.Results: A total of 238 uncomplicated TBAD patients were included in the study on exploring the timing of endovascular treatment,including 142 cases in the acute TEVAR group and 96 cases in the subacute TEVAR group.After IPTW adjustment,all baseline variables were not statistically different between the two groups.The 30-day mortality rate was 1.5% in the acute TEVAR group and 0% in the subacute TEVAR group,and the difference was not statistically significant(P=0.242).The 30-day adverse event rate was 16.8% in the acute TEVAR group and 6.9% in the subacute TEVAR group(P=0.130).After 5 years of follow-up,all-cause mortality(HR,1.50;95%CI,0.59-3.81;P=0.393)and aortic-related mortality(HR,1.11;95%CI,0.34-3.60;P=0.861)between the two groups)with no significant difference.Landmark analysis showed that there was a significant treatment effect interaction over time for the composite outcome(Pinteraction=0.010).The incidence of composite outcome events in the acute TEVAR group within 1 year was significantly higher than that in the subacute TEVAR group(HR,0.25;95% CI,0.08-0.79;P=0.021),but there was no significant statistical difference within 1-5 years after TEVAR(HR,1.25;95% CI,0.56-2.76;P=0.594).The exploratory stent oversizing study ultimately included 226 patients with uncomplicated TBAD.The mean proximal oversizing was 2.1 ± 1.5 % in the ≤5% oversizing group and 9.6 ± 4.1 % in the >5% oversizing group.There was no significant difference in the 30-day incidence of adverse events(12.4% vs.15.1%,P=0.583)and mortality(0.7% vs.0%,P=1.000)between the two groups.After 5 years of follow-up,there was no significant difference in the survival rate free from all-cause death(93.3% vs.92.3%,P=0.957)and aortic-related death(95.0% vs.96.7%,P=0.928)between the two groups.However,the cumulative incidence of RTAD in the >5% oversizing group was significantly higher than that in the ≤5% oversizing group(P=0.007),and all RTADs occurred within 1 year after TEVAR.Conclusions: After 5 years of follow-up,the timing of TEVAR and stent oversizing had no significant impact on all-cause death and aortic-related death in patients with uncomplicated TBAD.However,acute TEVAR was significantly associated with an increased rate of adverse complications within 1 year after TEVAR,and stent oversizing >5% was significantly associated with an increased risk of RTAD within 1 year after TEVAR.Therefore,for patients with uncomplicated TBAD,waiting until the subacute stage and selecting a stent oversizing of ≤5% may be the best choice for TEVAR treatment.Part III Research on intelligent segmentation and reconstruction of true and false lumens of TBAD based on deep learningObjective: Accurate and intelligent diagnosis,risk assessment,preoperative planning and prognosis analysis of aortic dissection are crucial to individualized diagnosis and treatment.Existing CTA image post-processing methods based on threshold rendering cannot accurately display single lumen and thrombosed false lumens,cannot perform three-dimensional diameter and volume measurements,cannot perform hemodynamic model construction,etc.,and rely on manual editing,which is time-consuming and labor-intensive.Based on the prior knowledge of imaging,this study aims to develop a deep learning method based on the flap attention mechanism that can integrate the information of the upper and lower layers of the dissected artery to realize the intelligent segmentation and reconstruction of the true lumen(TL)and false lumen(FL)of TBAD.Methods: The preoperative CTA image data of TBAD patients with clear diagnosis were collected from four large tertiary hospitals to construct a multicenter data set for model training and testing.The dataset contains cases with and without thrombosis of the false lumen.All data were manually annotated using the medical image processing software Mimics 21.0(Materialise,Leuven,Belgium).Each CTA data is annotated layer by layer as TL,FL,branch vessels(Branch vessels,BR)and background.When thrombus exists in FL,it is included in FL for annotation.A deep learning model named ADSeg is proposed based on the Py Torch framework to develop a unique flap attention mechanism for TBAD segmentation,and combine the cascaded network structure and two-step training strategy.The model is trained with monocentric data,and the segmentation performance and generalization ability of the model are tested on a multicentric dataset.The segmentation results are evaluated by Dice coefficient and boundary distance.Results: A total of 108 TBAD patients were included in the multicenter dataset constructed in this study,including 68 cases in the training set and the remaining 40 cases(10 cases in each center)for internal validation and external testing.The Dice coefficients of TL,FL,AO(whole aorta)and BR segmented by ADSeg were 91.1%,88.4%,93.2% and 84.9%,respectively.Compared with the classic U-Net algorithm,the Dice coefficients of ADSeg to segment TL,FL and BR increased by 6.4%(91.1 vs.84.7%),7.4%(88.4% vs.81.0%)and 2.4%(84.9% vs.82.5%);the boundary distances were reduced by 0.36mm(1.47 mm vs.1.83mm),0.38mm(2.10 mm vs.2.48mm)and 0.10mm(1.80 mm vs.1.90mm).Compared with the most accurate method published previously,the Dice coefficients of ADSeg segmenting TL and FL are increased by 4.1%(91.1% vs.87.0%)and 5.9%(88.4% vs.82.5%)respectively,and the boundary distance is reduced by 0.21 mm(1.47 mm vs.1.68mm)and 0.64mm(2.10 mm vs.2.74mm).Compared with the powerful cascaded U-Net+,in the absence of thrombus in the false lumen,the Dice coefficients of ADSeg segmented TL and FL increased by 1.2%(90.1% vs.88.9%)and 1.1%(90.2% vs.89.1%),in the case of false lumen thrombosis,the Dice coefficients of ADSeg segmentation of TL and FL increased by 1.4%(92.0% vs.90.6%)and 2.8%(86.6% vs.83.8%),respectively.In addition,the segmentation performance of ADSeg in different centers is better than other algorithms,and it shows good generalization ability among different centers.Conclusions: In this study,we propose a deep learning method based on flap attention mechanism and optimized cascade network structure and two-step training strategy,which has higher segmentation accuracy than existing true and false lumens segmentation methods and shows strong generalization on a multicenter test set.ADSeg is expected to provide value for accurate and intelligent diagnosis of TBAD,individualized treatment plan development,postoperative follow-up and prognosis prediction.