研究背景：

鼻部作为面部美学和功能的核心区域，其精确测量与三维重建对整形外科的术前规划、术中决策及术后随访具有重要意义。传统二维测量因视角偏差和深度信息缺失，难以满足精准诊疗需求。而三维扫描技术（如3dMD）虽精度高（误差<1 mm），但高昂的设备成本、场所依赖性以及操作复杂性限制了其在远程医疗和基层医疗机构中的大范围应用。近年来，计算机视觉和人工智能技术的飞速发展为面部重建带来了新机遇。3D形态可变模型（3D Morphable Model，3DMM）通过参数化建模生成面部几何结构，条件生成对抗网络（Conditional Generative Adversarial Networks，cGAN）以对抗训练优化复杂的图像数据，结合两者可实现从二维图像到三维模型的高效转换，尤其对于非受控条件（如光照不均、姿势偏差）的图像处理中展现巨大潜力。然而，这些技术在鼻部整形测量中的准确性、稳定性和临床适用性仍需系统验证。

鼻翼缺损的修复是鼻再造领域的一项复杂挑战，常由外伤、肿瘤切除或先天畸形引起。鼻唇沟皮瓣因其可靠的面动脉血供、丰富的真皮下血管网以及高度的组织灵活性，成为鼻翼再造的首选技术之一。然而，术后常见并发症如鼻翼臃肿、供区瘢痕挛缩及鼻尖与鼻翼形态协调性不足等问题显著影响了功能与美学效果的平衡。目前，相关研究在以下方面存在不足：一方面，缺乏对鼻唇沟皮瓣修复效果的系统化定量分析及各阶段形态变化规律的深入探讨；另一方面，传统的皮瓣减容术虽能部分改善鼻翼外形，但难以充分满足美学需求，且可能因过度修整损伤皮瓣血供，增加坏死或感染等并发症风险。因此，亟需开发更精准的评估方法和优化手术策略，以提升鼻翼再造的整体效果。

本研究首先对3DMM与cGAN技术在鼻整形中的应用进行验证性探讨和可靠性分析，其次在3DMM与cGAN技术基础上，对基于上蒂鼻唇沟皮瓣的鼻翼再造序列患者进行一系列的分析和探讨。

研究目的：

（1）以3dMD扫描技术为对照，通过三维分析对比，验证3DMM拟合与cGAN的面部重建技术在鼻部测量中的准确性；

（2）对3DMM拟合与cGAN面部重建技术基于非受控图像的重建表现进行测量分析，评估该重建策略在临床远程诊疗实际条件下的可靠性；

（3）通过三维重建和分析技术，量化上蒂鼻唇沟皮瓣修复鼻翼缺损的序列手术各阶段的对称度，探讨鼻面部形态变化规律，为手术优化提供定量的证据；

（4）通过三维形态分析技术，对鼻翼复合组织翻转瓣进行美学指标的定量测量，本研究旨在系统评估其在鼻翼再造序列中的应用效果。

方法：

（1）选取2022年12月至2024年12月在中国医学科学院整形外科医院鼻整形中心寻求隆鼻手术的健康求美者，共24例。分别收集患者三维扫描数据及标准化图像组，采用3DMM拟合重建系统（3DMM-FRS）和基于条件生成对抗网络的层次化表征重建系统（cGAN-HRS）对图像组进行面部重建。分别将两种面部重建方法的模型与三维扫描模型进行对比分析。

（2）选取2023年6月至2024年6月在中国医学科学院整形外科医院鼻整形与鼻再造中心接受自体肋软骨鼻综合整形的病例，共26例。采用3DMM-FRS和cGAN-HRS联合重建的策略对低鼻患者的术前标准化图像组和术后非受控图像组进行面部重建，计算鼻指数、鼻梁指数、鼻唇角等相关指数的改善值，并与标准化图像组进行对比分析。

（3）选取2020年12月至2023年12月就诊于中国医学科学院整形外科医院鼻整形中心以“单侧鼻翼缺损”或“鼻翼畸形”为主诊断的患者，共22例。采用基于CT数据的面部重建技术和3DMM-FRS / cGAN-HRS分别对上蒂鼻唇沟皮瓣鼻翼再造序列的各阶段测量并对比分析鼻面部对称度指数，通过组内纵向比较以探讨鼻翼再造的变化规律，组间横向比较以探讨3DMM-FRS / cGAN-HRS的重建可靠性。

（4）选取2022年1月至2024年12月就诊于中国医学科学院整形外科医院鼻整形中心以“单侧鼻翼缺损”或“鼻翼畸形”为主诊断，且第 III 期手术方案为鼻翼复合组织翻转瓣联合皮瓣减容术的患者，共12例。采用基于CT数据的面部重建技术对序列各阶段进行重建，以健侧鼻翼M₀为基准，对患侧鼻翼M₁进行三维匹配性分析（RMSE和单向Hausdorff距离）和角度测量（鼻尖角和鼻尖旋转角）。

结果：

（1）与3dMD立体摄影三维扫描面部模型相比，3DMM-FRS和cGAN-HRS的全局RMSE分别为1.44±0.14 mm和1.43±0.21 mm，鼻部RMSE为0.79±0.11 mm和0.84±0.16 mm，两组与对照组的误差无统计学差异（p>0.05）。cGAN-HRS的面中部区域的RMSE 0.92±0.20mm，MAE（0.77±0.13 mm），和MedAE（0.63±0.14 mm）优于3DMM-FRS（1.04±0.12mm，0.87±0.09mm，0.71±0.11mm）。3DMM-FRS的鼻部Hausdorff距离（1.93±0.64 mm）优于cGAN-HRS（2.33±0.78 mm），差异具有统计学意义。特征点偏移误差中，两组均出现颏前点和颏下点显著偏倚，其余14个特征点的偏倚总量均在2mm以内。23项测量参数中，两组测量参数误差均在1mm以内，ICC高于0.9的频率均为78.79%。两组的鼻小柱宽度比对照组高0.14±0.29mm和0.18±0.21mm，3DMM-FRS左右侧鼻孔短径偏高0.41±0.45mm和0.31±0.62mm，cGAN-HRS的左右侧鼻孔短径偏高0.53±0.83mm和0.37±0.65mm，差异有统计学意义。面积测量中，3DMM-FRS的左侧鼻唇沟区和左侧面中部外侧区面积比对照组降低8.55%和5.57%，cGAN-HRS组的右侧鼻翼面积比对照组降低5.49%差异具有统计学差异。

（2）在鼻指数改善值上，复合重建组（-7.36±7.74）与标准拍摄组（-6.01±6.84）相比，绝对值平均增加 1.36（p＜0.05）。在其余 5项参数的改善值结果中，均无统计学差异。6 项参数 ICC 均高于 0.9，超出一致性界限范围的样本率≤7.69%。

（3）基于上蒂鼻唇沟皮瓣的鼻翼再造序列患者中，鼻额角在鼻唇沟皮瓣转移及断蒂术后，相比术前分别下降了 3.20±0.68°（p＜0.01），随后经历皮瓣减容术后则未见显著差异（p＞0.05）。鼻翼弧长对称度和鼻翼厚度对称度在鼻唇沟皮瓣转移及断蒂术后均出现显著升高，表现为高于健侧的超量恢复，随后在序列后阶段中出现回降并趋近于 1，与健侧数值更接近。然而，鼻翼厚度的回降趋势更显著，回降幅度为 0.48±0.07（p 均＜0.01）。术前鼻翼面积因固有缺损，对称度为0.72±0.18。在经历皮瓣转移和断蒂术后，患侧鼻翼面积剧增至健侧的 2 倍以上（2.74±0.62， p＜0.01），鼻唇沟区的对称度相对小幅度地降低至0.77±0.12（p＜0.01）。在序列完成后，鼻翼面积对称度稳定在1.20±0.25，鼻唇沟区对称度在 0.82±0.08。虽然，面中部内侧的对称度在皮瓣转移和断蒂术后出现下降，但降幅较小（0.14±0.02，p＜0.01），但在序列的终点可恢复至与健侧相当。面中部外侧区域各阶段均未见显著差异。

（4）在I期手术（皮瓣转移）前后，RMSE与单向Hausdorff距离的差异无统计学意义。在II期手术（皮瓣断蒂术）后，单向Hausdorff距离降低 1.77±0.56mm（p＜0.05）。在III期手术（鼻翼复合组织翻转瓣转移术联合皮瓣减容术）后，RMSE 降低 0.39±0.23mm，单向 Hausdorff距离降低了 1.59±0.56mm（p＜0.05）。II 期和 III 期手术后，RMSE和单Hausdorff距离共分别下降了0.71±0.23mm 和3.35±0.56mm。序列后的鼻尖角，相比皮瓣转移后低9.22±3.08°（p＜0.05）。

结论：

（1）两组在角度及线性测量中均表现良好。在鼻部面积测量中，3DMM-FRS凭借灵活的交互方式表现出更高的准确性，而在面中部的重建效果中，cGAN-HRS依靠多层次的表征调整机制，呈现更优越的表现。

（2）3DMM-FRS / cGAN-HRS联合重建策略在应对鼻整形相关的混合数据集时具有较好的可靠性和稳定性。

（3）再造鼻翼的形态由早期的超量修复逐渐恢复至远期较高的对称度，而鼻唇沟区则由早期急剧下降的容量逐渐回升。3DMM-FRS / cGAN-HRS 在序列始末阶段可提供较准确的重建效果，在变异度较高的阶段有待进一步提高精度。

（4）本研究通过三维分析验证了鼻翼复合组织翻转瓣在鼻翼缺损再造中的有效性。该复合组织瓣具有良好的血供基础，并通过对再造鼻翼的组织再分配，同时实现供区减容和受区提升的效果，显著改善了患侧鼻翼和鼻尖形态。

Background：

The nasal region, central to facial aesthetics and functionality, plays a critical role in facial aesthetics. The precise measurement and 3D reconstruction are essential for achieving optimal outcomes. Traditional 2D measurement techniques, constrained by perspective distortion and lack of depth information, fail to meet the demands of precision medicine. Although 3D scanning technologies, such as 3dMD, offer high accuracy (error <1 mm), their widespread adoption in remote healthcare and primary medical facilities is limited by high equipment costs, site dependency, and operational complexity. Recent advancements in computer vision and artificial intelligence have opened new avenues for facial reconstruction. The 3D Morphable Model (3DMM) generates facial geometry through parametric modeling, while Conditional Generative Adversarial Networks (cGAN) optimize complex image data via adversarial training. Together, these technologies enable efficient conversion from 2D images to 3D models, demonstrating significant potential in processing uncontrolled conditions, such as uneven lighting and pose variations. However, their accuracy, stability, and clinical applicability in nasal measurements require systematic validation.

Repair of full-thickness nasal ala defects represents another challenge in plastic surgery. Nasolabial flaps, valued for their reliable blood supply and tissue flexibility, are widely used in nasal ala reconstruction. Nonetheless, issues such as postoperative bulkiness, donor-site scarring, and suboptimal nasal tip harmony remain unresolved. Current research lacks quantitative analysis of morphological changes across surgical stages, and traditional flap debulking techniques, while improving nasal ala contour, often compromise aesthetic outcomes and risk vascular complications due to potential blood supply damage.

This study aims to validate the application of 3DMM and cGAN technologies in nasal reconstruction through exploratory and reliability analyses, followed by a comprehensive evaluation of superiorly based nasolabial flap reconstruction sequences in nasal ala reconstruction, leveraging 3DMM and cGAN techniques.

Objectives：

（1）To validate the accuracy of 3DMM fitting and cGAN-based facial reconstruction technologies in nasal measurements, using 3dMD scanning as the reference standard through 3D analysis.

（2）To evaluate the reconstruction performance of 3DMM fitting and cGAN technologies using uncontrolled images, assessing their reliability in clinical telemedicine scenarios.

（3）To quantify the symmetry of nasal alar reconstruction with superiorly based nasolabial flaps across surgical stages using 3D reconstruction and analysis, elucidating nasal-facial morphological changes to provide quantitative evidence for surgical optimization.

（4）To quantitatively measure aesthetic parameters of the nasal alar composite tissue turnover flap through 3D morphometric analysis, systematically evaluating its efficacy in nasal alar reconstruction sequences.

Methods：

（1）From December 2022 to December 2024, 24 healthy individuals seeking rhinoplasty at the Rhinoplasty Center, Hospital of Plastic and Reconstructive Surgery, Chinese Academy of Medical Sciences, were enrolled. 3D scanning data and standardized image sets were collected. Facial reconstruction was performed using the 3DMM Fitting Reconstruction System (3DMM-FRS) and the cGAN-based Hierarchical Representation Reconstruction System (cGAN-HRS). The resulting models were compared with 3dMD scanning models for accuracy analysis.

（2）From June 2023 to June 2024, 26 patients undergoing autologous rib cartilage rhinoplasty were included. A combined 3DMM-FRS and cGAN-HRS reconstruction strategy was applied to preoperative standardized and postoperative uncontrolled image sets. Improvements in nasal index, nasal bridge index, nasolabial angle, and related parameters were calculated and compared with standardized image sets.

（3）From December 2020 to December 2023, 22 patients diagnosed with unilateral nasal alar defects or deformities at the Rhinoplasty Center were recruited. CT-based facial reconstruction and 3DMM-FRS/cGAN-HRS were used to measure and compare nasal-facial symmetry indices across stages of superiorly based nasolabial flap reconstruction. Longitudinal within-group comparisons elucidated morphological changes, while cross-group comparisons assessed the reliability of 3DMM-FRS/cGAN-HRS reconstruction.

（4）From January 2022 to December 2024, 12 patients with unilateral nasal alar defects or deformities, undergoing Stage III surgery with nasal alar composite tissue turnover flap combined with flap debulking, were included. CT-based facial reconstruction was performed across surgical stages. Using the contralateral ala (M0) as the reference, 3D matching analysis (RMSE and unidirectional Hausdorff distance) and angular measurements (nasal tip angle and nasal tip rotation angle) were conducted on the affected ala (M1).

Results：

（1）Compared to 3dMD stereophotogrammetry models, 3DMM-FRS and cGAN-HRS exhibited global RMSEs of 1.44±0.14 mm and 1.43±0.21 mm, respectively, and nasal RMSEs of 0.79±0.11 mm and 0.84±0.16 mm, with no statistically significant differences from the control group (p>0.05). In the midface region, cGAN-HRS outperformed 3DMM-FRS with RMSE (0.92±0.20 mm), MAE (0.77±0.13 mm), and MedAE (0.63±0.14 mm) compared to 1.04±0.12 mm, 0.87±0.09 mm, and 0.71±0.11 mm, respectively. The nasal Hausdorff distance for 3DMM-FRS (1.93±0.64 mm) was superior to cGAN-HRS (2.33±0.78 mm), with statistically significant differences. Landmark deviation analysis revealed significant biases at the pogonion and submental points for both methods, while biases for the remaining 14 landmarks were within 2 mm. Among 23 measurement parameters, errors for both methods were within 1 mm, with ICCs above 0.9 in 78.79% of cases. Columellar width was 0.14±0.29 mm and 0.18±0.21 mm higher than the control for 3DMM-FRS and cGAN-HRS, respectively. Nostril short-axis measurements were higher by 0.41±0.45 mm and 0.31±0.62 mm (3DMM-FRS) and 0.53±0.83 mm and 0.37±0.65 mm (cGAN-HRS), with significant differences. In area measurements, 3DMM-FRS showed 8.55% and 5.57% reductions in the left nasolabial groove and left midface lateral areas, while cGAN-HRS showed a 5.49% reduction in the right alar area, with significant differences.

（2）The nasal index improvement in the composite reconstruction group (-7.36±7.74) was 1.36 higher in absolute value compared to the standardized imaging group (-6.01±6.84) (p<0.05). No significant differences were observed in the other five parameters. All six parameters had ICCs above 0.9, with sample rates exceeding consistency limits ≤7.69%.（3）In patients undergoing superiorly based nasolabial flap reconstruction, the nasofrontal angle decreased by 3.20±0.68° (p<0.01) after flap transfer and pedicle division compared to preoperative values, with no significant change after debulking (p>0.05). Alar arc length and thickness symmetry significantly increased post-transfer and pedicle division, reflecting overcorrection relative to the contralateral side, but regressed toward 1 in the final stage, approaching contralateral values. Alar thickness regression was more pronounced, with a reduction of 0.48±0.07 (p<0.01). Preoperative alar area symmetry was 0.72±0.18 due to inherent defects. Post-transfer and pedicle division, the affected alar area increased to over twice the contralateral area (2.74±0.62, p<0.01), while nasolabial groove symmetry slightly decreased to 0.77±0.12 (p<0.01). After the sequence, alar area symmetry stabilized at 1.20±0.25, and nasolabial groove symmetry at 0.82±0.08. Midface medial symmetry decreased post-transfer and pedicle division (0.14±0.02, p<0.01) but recovered to contralateral levels by the sequence’s end. No significant differences were observed in the midface lateral region across stages.

（4）No significant differences in RMSE or unidirectional Hausdorff distance were observed before and after Stage I (flap transfer). After Stage II (pedicle division), the unidirectional Hausdorff distance decreased by 1.77±0.56 mm (p<0.05). After Stage III (composite tissue turnover flap combined with debulking), RMSE decreased by 0.39±0.23 mm, and unidirectional Hausdorff distance by 1.59±0.56 mm (p<0.05). Post-Stage II and III, RMSE and Hausdorff distance decreased by 0.71±0.23 mm and 3.35±0.56 mm, respectively. The nasal tip angle post-sequence was 9.22±3.08° lower than post-flap transfer (p<0.05).

Conclusion：

（1）Both methods performed well in angular and linear measurements. In nasal area measurements, 3DMM-FRS demonstrated higher accuracy due to its flexible interactive approach, while cGAN-HRS excelled in midface reconstruction owing to its multilayered representation adjustment mechanism.

（2）The combined 3DMM-FRS/cGAN-HRS reconstruction strategy exhibited robust reliability and stability when handling mixed datasets in rhinoplasty.

（3）Reconstructed nasal alae transitioned from initial overcorrection to high long-term symmetry, while the nasolabial groove recovered from early volume loss. 3DMM-FRS/cGAN-HRS provided accurate reconstruction at the sequence’s start and end but requires further precision enhancement in stages with high variability.

（4）This study validated the efficacy of the nasal alar composite tissue turnover flap in full-thickness defect reconstruction through 3D analysis. The flap’s robust vascular foundation, coupled with tissue redistribution, simultaneously achieved donor-site debulking and recipient-site enhancement, significantly improving affected alar and nasal tip morphology and patient satisfaction.