第一部分 PAD患者肌容量下降与血管狭窄的关系及影响因素研究

背景：外周动脉疾病（PAD）是一种以下肢动脉狭窄或闭塞为特征的常见老年性疾病，其导致的慢性缺血状态常引发肌肉萎缩和功能障碍。PAD患者中常出现肌肉萎缩，显著影响患者的运动能力、生活质量及长期预后，目前尚无研究系统探讨PAD患者肌容量变化特征与血管狭窄程度之间的定量关联。

目的：本研究旨在基于CT图像定量分析PAD患者的下肢及腰部肌容量特征，探讨影响肌容量的临床相关因素，明确肌容量减少与血管狭窄程度的关联，构建预测模型识别肌容量减少的高风险人群，为PAD患者的个体化管理提供依据。

方法：回顾性纳入2018年12月至2023年7月在北京医院接受下肢CTA检查的233例PAD住院患者。收集其临床基本资料、生化指标及ADL评估等信息。基于CT图像使用Slice-O-matic软件测量L3、L4、大腿及小腿肌肉相关参数，计算全身骨骼肌指数MI。基于下肢CTA评价节段动脉狭窄程度。以四分位间距对MI进行分组，比较组间差异；血管节段狭窄程度与肌肉相关参数进行Spearman相关性分析；将组间差异显著的变量纳入多元线性回归模型筛选肌容量的独立影响因素，并基于回归模型构建Nomogram评分系统。

结果：最低MI组（Q1组）患者年龄显著较大，BMI较低，CRP、NLR、PLR等炎症指标水平升高，eGFR下降，Fontaine Ⅳ期和Rutherford ≥5级比例显著增高。Spearman相关性分析显示PoA、SFA、PTA等下肢远端动脉狭窄评分与骨骼肌面积、密度及指数等多个指标呈显著负相关。多元线性回归分析结果表明，BMI（β = 4.05, p < 0.001）、年龄（β = -0.82, p < 0.01）Rutherford ≥5级（β = -31.60, p < 0.01）及PoA评分（β = -2.41, p < 0.05）是独立影响肌容量的关键因素。基于上述模型构建的Nomogram评分系统具有良好的预测能力，在区分高/低肌容量风险方面表现出显著效能。

结论：PAD患者的肌容量减少不仅与年龄和营养状态密切相关，更与下肢远端血管狭窄程度密切相关。构建的Nomogram评分工具可用于识别PAD患者中肌容量下降的高风险人群，具备临床风险分层及个体化干预的潜力。

第二部分 肌容量对外周动脉疾病预后的影响及预测模型构建

背景：外周动脉疾病（PAD）患者常面临全因死亡和大截肢等严重不良预后。近年来，骨骼肌减少被逐渐认为是影响PAD患者长期结局的重要因素之一。然而，当前尚缺乏将肌容量指标纳入PAD预后评估模型的系统研究。

目的：本研究旨在评估PAD患者肌容量与不良预后（死亡与截肢）之间的关系，并基于肌肉质量、动脉评分及炎症指标构建全因死亡与截肢的预测模型，为PAD个体化风险分层和干预策略提供依据。

方法：本研究纳入2018年12月至2023年7月在北京医院接受CTA检查的207例PAD住院患者，随访时间中位数为26.73个月。收集患者临床资料、肌肉CT评估参数、下肢动脉狭窄评分及实验室指标。以全因死亡和大截肢为主要结局，采用单因素与Lasso-Cox多因素回归分析，构建并验证预后模型，评估其判别力（C-index）与预测能力（time-ROC曲线）。

结果：共43例患者死亡，死亡组的肌容量指标（SMI、TMI、LMI、PMI等）显著低于生存组（P < 0.01），肌肉脂肪浸润指标（TFI、LFI）显著升高（P < 0.001）。Rutherford分级、舒张压、PLR等也与死亡相关。最终Lasso-Cox模型纳入MI指数、Rutherford分级、DBP与PLR，C-index为0.821，12月与24月AUC分别为0.897与0.891。共14例患者大截肢，截肢风险分析显示TMI和Rutherford为独立预测因素，截肢模型C-index为0.851。

结论：肌容量是PAD患者死亡与截肢风险的重要预测因子。将肌容量指标纳入PAD风险预测模型可提升预后评估的准确性。基于CT评估的全身骨骼肌指数（MI）可作为临床风险分层与干预管理的重要工具，具有良好的预测价值。

第三部分 基于V-Net的骨骼肌自动分割及其在外周动脉疾病预后预测中的应用

背景：外周动脉疾病（PAD）患者常出现肌容量和功能下降，严重影响其预后。在PAD患者中，CT估肌容量具有一定应用优势，但传统手工分割方法存在效率低、主观性强等局限。卷积神经网络（CNN）技术在医学图像分割领域表现出良好前景，有望实现肌容量评估的自动化与标准化。

目的：本研究旨在构建一套基于V-Net模型的腰椎CT图像自动分割系统，评估其分割性能，并探索自动提取的全身骨骼肌指数（MI）在PAD患者预后预测中的临床应用价值，同时建立性别特异的MI评估肌容量下降的cut-off值。

方法：收集242例PAD患者L3层面CT图像，采用3D Slicer进行人工勾画作为金标准，构建V-Net卷积神经网络模型进行训练与测试。评估指标包括Dice系数与HD95。将自动分割获得的MI值用于构建Cox生存回归模型并进行验证，绘制Kaplan-Meier曲线、timeROC曲线，并计算C-index与AUC。采用最大log-rank统计量法在不同性别中寻找最佳cut-off值。

结果：模型在独立测试集上的平均Dice系数为0.894，HD95为3.2±0.8mm，分割精度较高。基于自动MI的生存模型C-index为0.828，1年与2年时点的AUC分别为0.904与0.894。引入MI后模型拟合显著改善（χ² = 14.35，P < 0.001，AIC下降12.3）。MI在男性与女性中的最佳cut-off值分别为149.59 cm²/m²与91.57 cm²/m²，均能显著区分生存风险。

结论：V-Net模型可实现PAD患者腰椎CT图像骨骼肌的精准自动分割。自动获得的MI值可替代人工测量进行肌容量评估，并具有独立的预后预测能力。本研究还建立了基于预后的性别特异的MI的cut-off值，为PAD患者的精准诊疗提供了高效、可推广的辅助工具。

Part 1 : CT-Based Evaluation of Muscle Mass and Its Clinical Determinants in Patients with Peripheral Arterial Disease

Background: Peripheral arterial disease (PAD) is a prevalent atherosclerotic condition characterized by lower limb ischemia, commonly accompanied by muscle atrophy and impaired mobility. Muscle loss is a frequent comorbidity among PAD patients, contributing to poor functional status and adverse outcomes. However, the relationship between muscle mass decline and vascular stenosis in PAD remains unclear.

Objective: This study aims to quantitatively evaluate muscle characteristics in the lower limbs and lumbar region using CT imaging in PAD patients, identify clinical predictors of low muscle mass, and construct a nomogram to stratify muscle loss risk.

Methods: A total of 233 hospitalized PAD patients undergoing lower limb CTA at Beijing Hospital from December 2018 to July 2023 were retrospectively enrolled. Clinical data, laboratory parameters, vascular stenosis scores, and ADL assessments were collected. Muscle areas and densities at the L3, L4, thigh, and calf levels were measured via Slice-O-matic software, and a composite muscle index (SMI + TMI + LMI) was calculated. Univariate, correlation, and multivariate linear regression analyses were conducted to identify factors associated with muscle mass. A predictive nomogram was developed based on significant variables.

Results: Patients in the lowest muscle mass quartile (Q1) had significantly higher age, lower BMI, increased inflammatory markers (CRP, NLR, PLR), reduced eGFR, and more severe PAD staging (Fontaine IV and Rutherford ≥5). Spearman correlation revealed that stenosis scores of distal arteries (e.g., PoA, SFA, PTA) were negatively associated with muscle indices. In multivariate analysis, BMI (β = 4.05, p < 0.001), age (β = -0.82, p < 0.01), Rutherford ≥5 (β = -31.60, p < 0.01), and PoA stenosis score (β = -2.41, p < 0.05) were independent predictors of muscle mass. The resulting nomogram demonstrated good discrimination in stratifying high- and low-risk muscle loss groups.

Conclusions: Muscle loss in PAD patients is not only associated with aging and nutritional status but is also significantly influenced by the severity of lower limb arterial stenosis. The developed nomogram offers a practical tool for early identification of high-risk muscle loss in PAD and may inform individualized management strategies.

Part 2 : Construction of a Prognostic Model Based on Muscle Mass for Patients with Peripheral Arterial Disease

Background: Patients with peripheral arterial disease (PAD) are at high risk of adverse outcomes, including all-cause mortality and major lower limb amputation. Emerging evidence suggests that reduced skeletal muscle mass, or sarcopenia, plays a critical role in PAD prognosis, but studies integrating muscle metrics into predictive models remain limited.

Objective: This study aimed to investigate the relationship between muscle mass and poor outcomes in PAD patients and to develop and validate predictive models for all-cause mortality and amputation based on muscle indices, arterial stenosis scores, and inflammatory biomarkers.

Methods: A total of 207 hospitalized PAD patients undergoing CTA at Beijing Hospital from December 2018 to July 2023 were enrolled. Median follow-up was 26.73 months. Clinical, imaging, and laboratory data were collected. Cox regression and Lasso-Cox models were applied to identify predictors of mortality and amputation. Model performance was evaluated using the concordance index (C-index) and time-dependent ROC curves.

Results: Forty-three patients died during follow-up. Muscle indices including SMI, TMI, LMI, and PMI were significantly lower in the mortality group (P < 0.01), while fat infiltration indices (TFI, LFI) were higher (P < 0.001). Rutherford classification, diastolic blood pressure (DBP), and platelet-to-lymphocyte ratio (PLR) were also associated with mortality. The final Lasso-Cox model incorporated MI index, Rutherford stage, DBP, and PLR, achieving a C-index of 0.821 and AUCs of 0.897 and 0.891 at 12 and 24 months, respectively. For amputation prediction, TMI and Rutherford stage were identified as independent predictors, with a C-index of 0.851.

Conclusions: Reduced muscle mass and increased fat infiltration are strong predictors of mortality and amputation in PAD. Incorporating CT-derived muscle indices into prediction models enhances risk stratification. The muscle index (MI) shows strong clinical utility and prognostic value for guiding individualized management of PAD.

Part 3 ：Automatic Skeletal Muscle Segmentation Based on V-Net and Its Application in Prognostic Prediction for Peripheral Artery Disease

Background : Sarcopenia is a common comorbidity in patients with peripheral arterial disease (PAD) and is associated with poor prognosis. CT-based assessment at the L3 level is recognized as a gold standard for muscle mass quantification. However, manual segmentation is time-consuming and subject to interobserver variability. Convolutional neural networks (CNNs) have shown great potential in automating medical image analysis.

Objective : This study aimed to develop and validate an automatic skeletal muscle segmentation model based on V-Net for L3-level CT images in PAD patients. It further assessed the prognostic value of automatically extracted muscle index (MI) and established sex-specific cut-off values for sarcopenia diagnosis.

Methods : A total of 242 CT images from PAD patients were manually annotated and used to train a V-Net segmentation model. Performance was evaluated using Dice similarity coefficient and 95th percentile Hausdorff Distance (HD95). The MI derived from automatic segmentation was applied to a Cox regression model for survival prediction, validated by C-index, Kaplan-Meier curves, and time-dependent ROC analysis. Optimal MI cut-offs for men and women were determined using maximally selected log-rank statistics.

Results : The model achieved an average Dice coefficient of 0.894 and HD95 of 3.2±0.8 mm on the test set. The automatically obtained MI yielded a C-index of 0.828; AUCs were 0.904 at 1 year and 0.894 at 2 years. Adding MI to the clinical model significantly improved model fit (χ² = 14.35, P < 0.001; AIC reduced by 12.3). The optimal MI cut-off values were 149.59 cm²/m² for males and 91.57 cm²/m² for females, both stratifying survival effectively.

Conclusion : The V-Net model enables accurate automatic segmentation of lumbar CT-derived skeletal muscle in patients with peripheral artery disease (PAD). The automatically generated muscle index (MI) can serve as a reliable alternative to manual measurements for muscle mass evaluation and demonstrates independent prognostic value. Furthermore, this study established sex-specific, prognosis-based MI cut-off values, providing an efficient and generalizable tool to support precision management in PAD patients.