第一部分：我国中重度以上二尖瓣反流患者行经导管二尖瓣缘对缘修复适用性的解剖学分类及临床结局：一项真实世界的超声心动图筛查研究

背景：经导管二尖瓣缘对缘修复(Mitral valve transcatheter edge-to-edge repair，M-TEER)已成为手术禁忌或高危二尖瓣反流患者的二尖瓣反流(Mitral regurgitation, MR)的重要治疗选择，但关于亚洲人接受M-TEER治疗的数据十分有限。本研究旨在评估和总结我国中重度以上MR患者行M-TEER的解剖适用性及临床结局。

方法：本研究连续纳入2021年1月至2024年2月期间在阜外医院接受M-TEER超声心动图筛查的有症状中重度以上MR患者。所有患者均按现行M-TEER解剖学分类标准分类。总结所有患者的临床特征、超声心动图参数和最终治疗方案。重点分析在我院接受M-TEER治疗患者的临床结局，包括手术成功率和术后1年MR复发的累积发生率。

结果：本研究最终纳入481例中重度以上MR患者，其中包括304例退行性二尖瓣反流(Degenerative mitral regurgitation，DMR)和150例功能性二尖瓣反流(Functional mitral regurgitation，FMR)患者。平均年龄为68.81±10.06岁，44.28%为女性，70.69%为重度MR。全组患者平均左心房前后径为49.78±11.12 mm，平均左心室舒张末期内径为58.08±7.90 mm，平均左心室射血分数为58.82±13.33%。DMR患者左心房前后径(47.79±9.67 mm vs 53.41 ± 12.97 mm, P<.001)和左心室舒张末期内径(56.17±6.59 mm vs 61.21±9.06 mm, P<.001)显著小于FMR患者，左心室射血分数则更高(65.08±8.20% vs 47.45±13.52%, P<.001)。理想的、适合的、挑战性的和不可行的M-TEER解剖分别占全组患者的23.49%、36.17%、13.93%和26.4%。在DMR患者中，这些比例分别为24.67%、37.5%、12.5%和25.33%；在FMR患者中则分别为24.67%、32.67%、14%和28.67%。在接受M-TEER治疗的患者中，理想的、适合的、挑战性的和不可行的解剖分别占36%、46%、15%和3%。全组的手术成功率为91.88%，其中DMR和FMR患者间无显著差异(92.2% vs 91.07%, P=1)。理想/可行解剖患者的手术成功率显著高于挑战性/不可行解剖患者(95.81% vs 71.05%，P <.001)。全组平均随访18.72 ± 12.46个月，DMR和FMR患者的全因死亡率分别为5.93%和7.84%。术后90天和1年，DMR患者分别有83.5%和78.3%的患者MR程度为≤1+，而FMR患者分别为75%和62.1%。术后1年时，DMR和FMR患者中NYHA心功能分级为I~II级的比例分别为88.2%和81%。理想/可行解剖患者术后1年MR复发率显著低于挑战性/不可行解剖患者(5.28%vs 11.86%，P=0.003)。多变量竞争风险回归分析结果显示瓣叶-瓣环指数＜1.2的患者在M-TEER术后1年内MR复发的风险显著增加478%。

结论：中重度以上MR患者中59.66%为理想/适合的M-TEER解剖，但仅48%最终接受M-TEER治疗。无干预的患者中，有42%具有理想或适合的解剖条件，提示存在治疗不足。我国患者接受M-TEER的手术成功率与欧美国家相似，但仍需要进一步研究手术疗效是否存在族裔差异。不同MR病因在M-TEER手术结局方面无显著差异，而解剖理想/可行的患者接受M-TEER的手术成功率相比挑战性/不可行的更高，复发风险更低。

第二部分：经导管二尖瓣缘对缘修复治疗中央区与非中央区退行性二尖瓣反流的比较分析

目的：标志性的随机对照临床试验已证实经导管二尖瓣缘对缘修复术（Mitral Valve Transcatheter Edge-to-Edge Repair, M-TEER）治疗高危退行性二尖瓣反流（Degenerative Mitral Regurgitation, DMR）患者的安全性和有效性，但存在严格的解剖结构纳入排除标准。本文旨在评价临床试验纳入标准外的非中央区病变DMR患者相比中央区病变DMR患者的手术与预后差异。

方法：本研究回顾性分析了2021年1月至2024年2月在中国医学科学院阜外医院因药物难治性中重度DMR接受M-TEER治疗的患者临床资料。将患者根据病变部位分为中央区病变组和非中央区病变组。通过电子病历系统收集患者的人口统计学信息、临床表现、影像资料及围术期资料，并进行规律术后随访。通过Kaplan-Meier生存曲线分析出院时M-TEER手术成功患者术后1年和3年免于二尖瓣反流复发的生存率以及Cox比例风险回归分析与二尖瓣反流复发或死亡相关的危险因素。

结果：本研究共纳入136例患者，其中中央区病变组77例，非中央区病变组59例。中央区病变组患者的左心室射血分数显著低于非中央区病变组（64.68±7.16% vs 67.07± 5.09%, P=0.031）。两组患者在其他基线特征和术前超声心动图参数方面无显著差异。两组手术成功率相似（中央区病变组：93.51% vs 非中央区病变组：91.53%，P=0.92）。非中央区病变组患者中59.32%植入1枚夹子，而中央区病变组患者中50.65%植入≥2枚夹子。两组在术后并发症发生率方面无显著差异。中央区病变组有1例住院死亡，1例单瓣叶夹持，3例心包积液；非中央区病变组有2例单瓣叶夹持，3例术中中转外科手术。

术后90天及1年随访显示，两组患者的二尖瓣反流程度均显著减轻。中央区病变组和非中央区病变组患者术后1年内残余反流≤1+的比例分别为80%和80.56%，术后左心房和左心室尺寸均显著减少，提示心脏逆重构。出院时非中央区病变组的理想M-TEER手术成功率高于中央区病变组（83.05% vs. 71.43%，P=0.11），但差异无统计学意义。Kaplan-Meier生存曲线分析显示，非中央区病变组患者术后3年免于二尖瓣反流复发的生存率和中央区病变组相似（94.9% vs. 90.3%，P=0.46）。多因素Cox比例风险回归分析显示，出院时二尖瓣平均跨瓣压差每增加1 mmHg，出现二尖瓣反流复发或死亡的风险增加36%（HR=1.36，95% CI：1.121.66，P=0.002）；瓣叶-瓣环指数≤1.2的患者发生二尖瓣反流复发或死亡的风险是瓣叶-瓣环指数>1.2患者的3.46倍（95% CI：1.0011.99，P=0.05）。

结论：经选择的非中央区MR患者行M-TEER可以达到与中央区MR患者相似的手术成功率而不增加手术并发症发生率。相比中央区DMR患者，非中央区MR患者行M-TEER对瓣口面积影响更小，达到理想的M-TEER手术成功率和术后免于二尖瓣反流复发的生存率似乎更高。此外，出院时二尖瓣平均跨瓣压差和瓣叶-瓣环指数≤1.2是M-TEER术后二尖瓣反流复发或死亡的独立危险因素。

第三部分：基于深度学习的心脏计算机断层扫描自动二尖瓣装置分割与测量模型的开发与验证

目的：多层螺旋计算机断层扫描（MSCT）是经导管二尖瓣置换术（Transcatheter Mitral Valve Replacement, TMVR）术前影像学评估的关键工具。然而目前的二尖瓣测量主要依赖影像专家手动评估，过程复杂且主观性强。因此，本研究旨在开发并验证一种基于深度学习的自动二尖瓣装置分割与测量模型，以提高影像分析效率并减少人为误差。

方法：本研究纳入2021年1月至2023年12月期间在阜外医院就诊并接受TMVR筛查的162例退行性二尖瓣反流（DMR）患者，以及403例健康对照者，共565例患者。所有受试者均接受冠状动脉计算机断层成像血管造影（CCTA）检查，并通过3mensio软件手动测量二尖瓣环参数，以作为真实值对比。研究采用MedNeXt网络架构，结合改进的卷积神经网络进行自动分割，并引入解剖学约束和轮廓损失函数优化分割精度和边界识别能力。最终，通过Dice系数、Hausdorff距离（HD95）和结构距离相似性（SDC）评估模型分割性能，并采用相关性分析和Bland-Altman分析对比自动测量与手动测量的一致性。

结果：本研究共纳入275,800张CCTA图像，并采用分层随机抽样将数据集按75%:15%:15%的比例划分为训练集（114例DMR患者、282例正常对照者，共193,060张图像）、测试集（24例DMR患者、60例正常对照者，共43,220张图像）和验证集（24例DMR患者、61例正常对照者，共39,520张图像）。整个队列的平均年龄为60.91±11.31岁，女性占比44.25%，其中高血压（46.90%）和高血脂（38.76%）为常见合并症。整体而言，DMR患者与正常对照者在训练集、测试集和验证集中的分布均衡，基线特征及大部分超声心动图参数差异无统计学意义。模型在左心房、左心室等大器官的Dice系数达0.95以上，二尖瓣前叶和后叶的Dice系数分别为0.874和0.864，表明其对瓣叶形态的良好识别能力。在自动测量与手动测量对比分析中，瓣环面积、瓣环周长、TT径、CC径和AP径的相关性均超过0.90（P < 0.001），Bland-Altman分析显示，超过90%的测量数据落在一致性限（LoA）范围内，验证了自动测量的可靠性。然而，瓣环非平面化角度和乳头肌至瓣环平面距离的相关性较低（r < 0.50），提示自动测量在瓣膜复杂几何形态评估方面仍有优化空间。此外，舒张末期的测量稳定性低于收缩末期，舒张末期瓣环周长的一致性限（LoA）范围显著宽于收缩末期（51.98 vs 14.18），舒张末期瓣环周长的相关性（r = 0.62）低于收缩末期（r = 0.97），表明舒张末期测量受解剖变异影响较大，测量精度较低。

结论：本研究基于深度学习开发并验证了一种高效的MSCT二尖瓣装置自动分割与测量模型。该模型在瓣环面积、周长及径向测量上与手动测量高度一致，整体临床可行性较高，但舒张末期测量稳定性低于收缩末期。

Section 1: Anatomical Suitability and Clinical Outcomes of Transcatheter Edge-to-Edge Repair in Chinese Patients with Significant Mitral Regurgitation: A Real-World Echocardiographic Screening Study

Objectives: Mitral valve transcatheter edge-to-edge repair (M-TEER) has become a crucial treatment option for patients with mitral regurgitation (MR) who are either inoperable or at high surgical risk. However, data on Asian patients undergoing M-TEER are very limited. This study aims to evaluate the anatomical suitability and clinical efficacy of M-TEER in patients with moderate to severe or severe MR in China.

Methods: This study consecutively enrolled symptomatic patients with moderate-to-severe or severe MR who underwent M-TEER echocardiographic screening at Fuwai Hospital from January 2021 to February 2024. All patients were classified according to the current anatomical classification criteria for M-TEER. Clinical characteristics, echocardiographic parameters, and final treatment plans were summarized for all patients. The clinical outcomes of patients undergoing M-TEER at our hospital were analyzed, focusing on the procedural success rate and the cumulative incidence of MR recurrence one year postoperatively.

Results: A total of 481 patients with moderate to severe MR were included in the study, comprising 304 patients with degenerative mitral regurgitation (DMR) and 150 patients with functional mitral regurgitation (FMR). The average age was 68.81±10.06 years, 44.28% were female, and 70.69% had severe MR. The average anteroposterior diameter of the left atrium was 49.78±11.12 mm, the average end-diastolic diameter of the left ventricle was 58.08±7.90 mm, and the average left ventricular ejection fraction was 58.82±13.33%. The anteroposterior diameter of the left atrium (47.79±9.67 mm vs. 53.41±12.97 mm, P<.001) and the end-diastolic diameter of the left ventricle (56.17±6.59 mm vs. 61.21±9.06 mm, P<.001) were significantly smaller in DMR patients than in FMR patients, while the left ventricular ejection fraction was higher in DMR patients (65.08±8.20% vs. 47.45±13.52%, P<.001). The anatomical suitability for M-TEER was classified as ideal, suitable, challenging, and impossible in 23.49%, 36.17%, 13.93%, and 26.4% of patients, respectively. In DMR patients, these proportions were 24.67%, 37.5%, 12.5%, and 25.33%, while in FMR patients, they were 24.67%, 32.67%, 14%, and 28.67%. Among patients who underwent M-TEER, the proportions of ideal, suitable, challenging, and impossible anatomy were 36%, 46%, 15%, and 3%, respectively.  The overall procedural success rate was 91.88%, with no significant difference between DMR and FMR patients (92.2% vs. 91.07%, P=1). The procedural success rate was significantly higher in patients with ideal/feasible anatomy compared to those with challenging/impossible anatomy (95.81% vs. 71.05%, P<.001). The mean follow-up duration for the entire cohort was 18.72 ± 12.46 months, with all-cause mortality rates of 5.93% and 7.84% in DMR and FMR patients, respectively. At 90 days and one year postoperatively, 83.5% and 78.3% of DMR patients had MR grades of ≤1+, compared to 75% and 62.1% of FMR patients, respectively. One year after the procedure, the proportions of DMR and FMR patients with NYHA class I-II heart function were 88.2% and 81%, respectively. The one-year postoperative MR recurrence rate was significantly lower in patients with ideal/feasible anatomy compared to those with challenging/impossible anatomy (5.28% vs. 11.86%, P=0.003). Multivariable competing risk regression analysis showed that patients with a leaflet-to-annulus index of <1.2 had a 478% increased risk of MR recurrence within one-year post-M-TEER.

Conclusion: Among patients with moderate-to-severe or severe MR, 59.66% had ideal/ suitable anatomy for M-TEER, but only 48% ultimately underwent M-TEER. Among untreated patients, 42% had ideal or suitable anatomy, indicating potential undertreatment. The procedural success rate of M-TEER in Chinese patients is similar to that in Western countries, but further studies are needed to determine whether there are ethnic differences in procedural outcomes. There were no significant differences in procedural outcomes between different MR etiologies, but patients with ideal/suitable anatomy had higher procedural success rates and lower MR recurrence risks compared to those with challenging/impossible anatomy.

Section 2: Comparative Analysis of Transcatheter Edge-to-Edge Mitral Valve Repair for Central vs. Non-central Degenerative Mitral Regurgitation

Objective: Landmark randomized controlled trials have demonstrated the safety and efficacy of Mitral Valve Transcatheter Edge-to-Edge Repair (M-TEER) in high-risk patients with degenerative mitral regurgitation (DMR). However, these studies applied strict anatomical eligibility criteria. This study aims to evaluate the differences in procedural outcomes and prognosis between patients with non-central DMR, who were excluded by clinical trial standards, and those with central DMR lesions.

Methods: This retrospective study analyzed the clinical data of patients with drug-resistant moderate to severe DMR who underwent M-TEER at Fuwai Hospital from January 2021 to February 2024. Patients were categorized into two groups based on lesion location: central and non-central DMR. Demographic data, clinical presentations, imaging data, and perioperative records were obtained from electronic medical records, with regular postoperative follow-ups conducted. Kaplan-Meier survival curves were utilized to analyze the 1-year and 3-year survival rates free from mitral regurgitation (MR) recurrence in patients who had successful M-TEER at discharge. Cox proportional hazards regression analysis was performed to identify risk factors associated with MR recurrence or mortality.

Results: A total of 136 patients were included in the study, with 77 in the central DMR group and 59 in the non-central DMR group. The left ventricular ejection fraction (LVEF) was significantly lower in the central DMR group compared to the non-central group (64.68±7.16% vs. 67.07±5.09%, P=0.031). No significant differences were observed in other baseline characteristics or preoperative echocardiographic parameters between the two groups. The procedural success rates were similar (93.51% in the central group vs. 91.53% in the non-central group, P=0.92). In the non-central DMR group, 59.32% of patients had one clip implanted, whereas 50.65% of patients in the central group had two or more clips implanted. Postoperative complication rates did not differ significantly between the groups. One patient in the central group experienced in-hospital mortality, along with one case of single-leaflet device attachment (SLDA) and three cases of pericardial effusion. In the non-central group, two cases of SLDA and three intraoperative conversions to surgery were reported.

At 90-day and 1-year follow-up, both groups showed a significant reduction in MR severity. At 1 year postoperatively, 80% of patients in the central group and 80.56% in the non-central group had residual MR ≤1+. Significant reductions in left atrial and ventricular sizes were also observed, indicating reverse cardiac remodeling. The ideal M-TEER procedural success rate was higher in the non-central group compared to the central group at discharge (83.05% vs. 71.43%, P=0.11), although this difference was not statistically significant. Kaplan-Meier survival analysis revealed similar 3-year survival rates free from MR recurrence between the non-central and central groups (94.9% vs. 90.3%, P=0.46). Cox proportional hazards regression showed that each 1 mmHg increase in the mean mitral valve pressure gradient at discharge was associated with a 36% higher risk of MR recurrence or death (HR=1.36, 95% CI: 1.12–1.66, P=0.002). Furthermore, patients with a leaflet-to-annulus index ≤1.2 were 3.46 times more likely to experience MR recurrence or death compared to those with an index >1.2 (95% CI: 1.00–11.99, P=0.05).

Conclusion: M-TEER in selected non-central DMR patients achieves similar procedural success rates as in central DMR patients, without an increased risk of complications. Non-central DMR patients exhibited less reduction in mitral valve orifice area compared to central DMR patients and demonstrated higher ideal M-TEER procedural success rates and better survival free from MR recurrence. Additionally, the mean mitral valve pressure gradient at discharge and a leaflet-to-annulus index ≤1.2 were identified as independent risk factors for MR recurrence or death following M-TEER.

Section 3: Development and Validation of a Deep Learning-Based Automatic Segmentation and Measurement Model for Mitral Valve Apparatus in Cardiac Computed Tomography

Objective: Multi-slice computed tomography (MSCT) is pivotal for preprocedural imaging evaluation in transcatheter mitral valve replacement (TMVR). However, current mitral valve measurements rely on time-consuming and subjective manual assessments by imaging experts. This study aimed to develop and validate a deep learning-based automated mitral valve apparatus segmentation and measurement model to enhance imaging analysis efficiency and reduce human error.

Methods: A total of 565 subjects were enrolled, including 162 patients with degenerative mitral regurgitation (DMR) undergoing TMVR screening and 403 healthy controls at Fuwai Hospital from January 2021 to December 2023. All subjects underwent coronary computed tomography angiography (CCTA), and mitral annular parameters were manually measured using 3mensio software as ground truth. An improved MedNeXt network architecture was employed for automated segmentation, integrating anatomical constraints and contour loss functions to optimize segmentation accuracy and boundary recognition. Model performance was evaluated using Dice coefficients, 95% Hausdorff distance (HD95), and structural distance similarity (SDC). Consistency between automated and manual measurements was assessed via correlation and Bland-Altman analyses.

Results: A total of 275,800 CCTA images were analyzed. The dataset was stratified into training (114 DMR patients, 282 controls; 193,060 images, 75%), testing (24 DMR patients, 60 controls; 43,220 images, 15%), and validation sets (24 DMR patients, 61 controls; 39,520 images, 15%). The overall cohort had a mean age of 60.91 ± 11.31 years, with 44.25% female patients. Hypertension (46.90%) and hyperlipidemia (38.76%) were the most common comorbidities. The distribution of DMR patients and healthy controls was balanced across datasets, with no significant differences in baseline characteristics or most echocardiographic parameters. The model achieved Dice coefficients >0.95 for large cardiac structures (left atrium, left ventricle), and 0.874/0.864 for anterior/posterior mitral leaflets, demonstrating robust morphological recognition. Automated measurements showed strong correlations with manual assessments for annular area, perimeter, TT distance, CC diameter, and AP diameter (r>0.90, P<0.001). Bland-Altman analysis showed that over 90% of measurements fell within the limits of agreement (LoA), demonstrating the reliability of the automatic measurements. However, correlations for annular non-planar angle (r<0.50) and papillary muscle-to-annulus distances (r<0.50) were suboptimal. Diastolic-phase measurements exhibited lower stability than systolic-phase results, with wider agreement limits (51.98 vs. 14.18) and reduced correlation (r=0.62 vs. r=0.97) for annular perimeter. This indicates that diastolic measurements are more susceptible to anatomical variations, leading to reduced accuracy.

Conclusion: This study developed and validated a deep learning-based automated mitral valve apparatus segmentation and measurement model using CCTA. The model demonstrated high accuracy in measuring mitral annulus parameters, achieving strong consistency with manual measurements, particularly for annular area, circumference, and diameters (r > 0.90). However, reduced stability in diastolic-phase measurements highlights the need for anatomical constraint optimization. Overall, this model exhibits strong clinical feasibility and has the potential to enhance preoperative TMVR assessment by improving the efficiency and consistency of imaging analysis.