Evolution of Facial Measurement Technology and Its Prospects with the Development of Artificial Intelligence
-
摘要:
面容测量在临床诊断和异常面容识别中具有重要意义。随着人体测量学的发展, 面容测量学已成为独立的研究领域, 并广泛应用于整形外科和颅颌面外科等学科。本文梳理了面容测量学的发展历程, 并探讨在人工智能背景下, 面容测量学的未来发展趋势。目前, 3D体表成像技术可准确捕捉和重建面部软组织的立体形态, 提高测定的精确性和客观性, 成为新的面容测量金标准, 不仅为疾病诊断和手术规划提供了参考, 且在美容效果评价和衰老研究中发挥重要作用。近年来人工智能技术发展迅猛, 可实现对异常面容的直接识别。基于二维图像的面容识别系统已相对成熟, 但受限于信息维度, 难以全面捕捉面部特征; 基于三维图像识别的准确度虽高, 却受限于训练样本数量, 在异常面容的识别与分类中仍面临挑战。人工智能与面容测量学的结合有效推动了面部标记点自动识别技术的发展, 为疾病面容评估提供了更为精确和可解释的方法。未来研究应聚焦于构建可靠的三维面容数据库, 以进一步提升面容识别的准确性; 同时, 应开发基于小样本的面容识别体系, 从而为罕见病和特殊疾病的面容识别提供有力支持。
Abstract:Facial anthropometry has profound importance in clinical diagnosis and the recognition of abnormal facial features. With the development of anthropometry, facial anthropometry has emerged as an independent research field and is widely applied in disciplines such as plastic surgery and cranio-maxillofacial surgery. This paper reviews the evolution of facial anthropometry and discusses its future trends in the context of artificial intelligence (AI). Currently, 3D facial imaging technology can accurately capture and reconstruct the three-dimensional morphology of facial soft tissues, and enhance the precision and objectivity of measurements, thus becoming the new "gold standard" for facial anthropometry. It not only provides reference for disease diagnosis and surgical planning but also plays a crucial role in evaluating cosmetic outcomes and aging research. In recent years, AI technology has developed rapidly, enabling direct recognition of abnormal facial features. Although facial recognition systems based on two-dimensional images are relatively mature, they have to struggle to fully capture facial features as they are limited by the dimensionality of information. While three-dimensional image-based recognition boasts high accuracy, it faces challenges in the recognition and classification of abnormal facial features due to limitations in the number of training samples. The integration of AI and facial anthropometry has effectively promoted automatic recognition technology for facial landmarks, thus providing more precise and interpretable methods for assessing disease-related facial features. Future research should focus on building reliable three-dimensional facial databases to further improve the accuracy of facial recognition. Additionally, developing facial recognition systems based on small sample sizes is necessary to provide robust support for the recognition of facial features associated with rare and special diseases.
-
Keywords:
- human phenotype /
- facial morphometry /
- facial recognition /
- facial diagnosis
-
系统性红斑狼疮(systemic lupus erythematosus, SLE)是一种多器官、多系统受累的慢性、炎症性、自身免疫性疾病[1]。随着诊疗技术的不断提高以及生物制剂的使用,目前SLE患者的10年生存率已超过90%[2]。但与此同时,慢性肾功能不全、心脑血管疾病以及肌肉骨骼等慢性并发症的发病率也逐年升高。股骨头缺血性坏死(avascular necrosis of femoral head, AVNF)是SLE肌肉骨骼病变中最值得关注的并发症之一[3],研究显示,4%~40%(平均约10%)的SLE患者可出现AVNF[1, 4-7]。发展至终末期时,药物治疗、髓芯减压术往往难以取得满意的疗效,严重影响患者的生存质量,最终约6% 的SLE-AVNF患者符合全髋关节置换术(total hip arthroplasty, THA)指征并进行手术治疗[8]。
THA是SLE并发终末期AVNF的重要治疗手段,其在缓解疼痛、改善功能及患者生活质量方面已得到大量研究证实[5-6, 9-11],预计到2030年,行THA的SLE-AVNF患者将比2005年增长174%[12]。然而,由于SLE患者通常多系统受累并长期使用免疫抑制剂,THA术后具有全身感染、切口愈合不良、SLE病情活动等潜在风险[2-5, 11, 13-14],且发生率高于一般患者[2, 5-6]。既往研究发现,术后早期是下肢关节置换术后发生严重并发症的主要阶段[5, 10, 15-16],因此,对THA合并SLE的早期安全性研究具有重要意义。本研究旨在通过回顾性分析,探究SLE患者行THA术后30 d内的并发症发生情况,以期为临床诊疗提供参考和借鉴。
1. 资料与方法
1.1 研究对象
本研究为回顾性队列研究,连续纳入2012年6月—2024年4月于北京协和医院骨科行THA患者的资料,根据手术患者是否伴有SLE分为SLE组和对照组。SLE组纳入标准:(1)根据1997年美国风湿病学会制定的分类标准确诊为SLE;(2)因AVNF接受THA治疗;(3)术后30 d随访资料完整。排除标准:(1)有恶性肿瘤疾病史;(2)术前30 d接受其他骨科手术;(3)急诊入院; (4)患有类风湿性关节炎、强直性脊柱炎、银屑病性关节炎等累及关节的风湿性疾病。对照组纳入标准:因创伤、发育性髋关节发育不良、酒精性股骨头坏死等原因接受THA手术。排除标准:(1)有恶性肿瘤疾病史;(2)术前30 d接受其他骨科手术;(3)急诊入院;(4)患有类风湿性关节炎、强直性脊柱炎、银屑病性关节炎等累及关节的风湿性疾病。
本研究已通过北京协和医院伦理审查委员会审批(审批号:K24C4121),并豁免患者知情同意。
1.2 围术期管理
1.2.1 术前
所有患者入院后完善血常规、肝/肾功能、血糖、电解质、凝血功能、红细胞沉降率、C反应蛋白、补体、抗双链DNA抗体、抗核抗体、尿常规、24 h尿蛋白等实验室检测以及髋关节X线等影像学检查。对于SLE患者,术前由风湿免疫科医生会诊,通过SLE疾病活动度评分(SLE disease activity index, SLEDAI)评估SLE活动情况。根据美国风湿病学会(American College of Rheumatology, ACR)/美国髋关节和膝关节外科协会指南[9],停用硫唑嘌呤、托法替布、环孢素、吗替麦考酚酯等免疫抑制剂及生物制剂,视病情保留或停用甲氨蝶呤、羟氯喹、他克莫司、来氟米特等免疫抑制剂,有激素使用史者将口服激素(如甲泼尼龙或泼尼松)调整为静脉滴注氢化可的松。对于合并其他疾病的患者积极作相应处理,例如,合并心脏病的患者完善心电图和超声心动图检查,合并脑血管疾病的患者完善磁共振血管成像检查。术前由麻醉科医生根据美国麻醉医师协会(American society of Anesthesiologists,ASA)分级评估患者状态,确保患者可耐受手术。
1.2.2 术中
在确认患者信息无误后,通过静脉滴注给予头孢呋辛。全身麻醉后,患者取侧卧位并固定骨盆。对术区进行常规消毒、铺巾和皮肤保护。采用髋部后外侧切口,依次切开、分离皮肤、皮下组织及髂肌后群,显露关节囊后予以切开,进行手法脱位。测量后截断股骨颈并取出股骨头。切除病变滑膜、增生的关节囊、髋臼盂唇、臼缘骨赘及纤维组织,使用两枚克氏针辅助固定,充分显露卵圆窝、髋臼底及骨性臼缘。依次锉磨髋臼,将金属髋臼杯以适当的前倾角和外展角植入髋臼进行试模。确保髋臼杯固定牢固后,安装陶瓷内衬并固定牢靠。屈曲、内收并内旋髋关节,充分显露股骨颈截骨面,在偏后外侧开髓。通髓腔后,依次使用近端髓腔锉扩髓,安放股骨颈试模并测量颈长。复位后测试髋关节的张力、活动度、稳定性和下肢长度。结果满意后,植入同型号股骨柄假体并打牢,随后安装同型号股骨头假体,复位髋关节,再次测试髋关节的活动度、稳定性、张力及下肢长度。结果满意后,用大量生理盐水冲洗伤口,并用含阿米卡星的生理盐水冲洗关节腔后吸净液体,彻底止血,再用氨甲环酸盐水浸泡创面。缝合重建髂肌后群,逐层缝合关闭切口。术中持续进行自体血回输,手术结束时不常规放置引流管。
1.2.3 术后
术后第1、3、5天常规行血常规、红细胞沉降率、C反应蛋白、肝/肾功能、凝血等实验室检测并观察切口愈合情况,血红蛋白低于70 g/L的患者行同种异体红细胞输注治疗。术后常规预防性使用抗生素3 d,口服抗凝药利伐沙班、镇痛药塞来昔布/艾瑞昔布2周。对于SLE患者,遵风湿免疫科会诊意见,术后1~3 d将静脉滴注氢化可的松调整为术前口服激素方案,术后1~2周视SLE活动程度、有无感染征象、切口愈合程度等情况恢复免疫抑制剂。术后第1天即鼓励患者拄拐下床活动,嘱咐患者站立、行走时注意肢体位置,谨防脱位和跌倒。
1.3 观察指标
以主要并发症(局部并发症和系统并发症)以及同种异体红细胞输注情况作为结局指标。局部并发症包括切口不良事件、假体周围感染、假体松动、假体周围骨折等。系统并发症包括深静脉血栓形成、肺血栓栓塞症、急性心血管事件、脑血管事件、肺炎、肾功能损伤、肝功能损伤、泌尿系统感染、SLE病情活动等。
1.4 偏倚控制
本研究在研究设计、数据处理过程中采取以下方法控制偏倚:首先,依据明确的纳入和排除标准筛选研究对象;其次,通过倾向性评分匹配研究对象,以最小化混杂因素的影响;最后,数值型数据由一名研究者独立录入,并由另一名研究者校对,影像学资料则由两名研究者共同评估并达成一致意见,确保所有指标的准确性和可靠性。
1.5 样本量估算
根据既往国内一项单中心研究[5],该研究具有良好的匹配度,SLE组术后主要并发症发生率为23%,而对照组为7%。采用PASS 2021软件进行样本量估算,设定显著性水平α=0.05,检验效能1-β=0.9,计算得出每组所需最低样本量为103例。
1.6 统计学处理
采用SPSS 26.0软件进行统计学分析。计量资料采用Shapiro-Wilk检验评估数据是否符合正态分布,年龄、体质量指数(body mass index,BMI)等符合正态分布的计量资料以均数±标准差表示,组间比较采用独立样本t检验;SLEDAI评分、C反应蛋白、D-二聚体等符合偏态分布的计量资料以中位数(四分位数)表示,组间比较采用非参数检验。计数资料以频数(百分数)表示,组间比较采用卡方检验或Fisher确切概率法。采用倾向性评分法对SLE组和对照组患者进行性别、年龄、手术侧别1∶1匹配,匹配容差设为0.02,比较两组主要并发症以及同种异体红细胞输注情况。双侧检验,以P<0.05为差异具有统计学意义。
2. 结果
2.1 SLE组患者基本情况
根据纳入和排除标准,研究组共入选270例SLE患者,平均年龄(37.8±12.1)岁,SLE病程为8.7(4.0, 12.0)年,SLEDAI评分为0.16(0, 0)。围术期糖皮质激素使用率为77.04%(表 1)。18例(6.67%)患者在术后30 d内出现主要急性并发症,其中2例(0.74%)患者发生上呼吸道感染;2例(0.74%,其中1例发生感染性休克)患者发生肺部感染;3例(1.11%)患者出现泌尿系感染;2例(0.74%)患者出现其他系统感染;1例(0.37%) 患者应激性溃疡伴出血;5例(1.85%)患者发生切口愈合不良;1例(0.37%)患者发生切口感染;1例(0.37%)患者输注血小板后出现休克;1例(0.37%) 患者出现SLE病情活动。所有患者经积极保守治疗后均好转出院。61例(22.59%)患者接受了同种异体红细胞输注。
表 1 270例行THA术的SLE患者基本特征Table 1. Fundamental characteristics of 270 cases of patients with SLE who underwent THA surgery指标 数值 年龄(x±s,岁) 37.8±12.1 性别[n(%)] 男性 41(15.19) 女性 229(84.81) BMI(x±s,kg/m2) 22.9±3.8 ASA分级[n(%)] ≤Ⅱ级 239(88.52) >Ⅱ级 31(11.48) 手术侧别[n(%)] 单侧 194(71.85) 双侧 76(28.15) 手术时间[M(P25, P75), min] 126.3(85.0, 170.0) 术中出血[M(P25, P75), mL] 333.6(200.0, 400.0) 合并症[n(%)] 骨质疏松 64(23.70) 高血压 37(13.70) 冠心病 3(1.11) 糖尿病 5(1.85) 脑血管疾病 3(1.11) 呼吸系统疾病 16(5.93) 消化系统疾病 11(4.07) 泌尿系统疾病 41(15.19) 血液系统疾病 4(1.48) 继发或伴发其他风湿性疾病 17(6.30) SLE病程[M(P25, P75),年] 8.7(4.0, 12.0) SLEDAI评分[M(P25, P75),分] 0.16(0, 0) SLE围术期用药[n(%)] 羟氯喹 193(71.48) 来氟米特 21(7.78) 他克莫司 16(5.93) 甲氨蝶呤 18(6.67) 氢化可的松 208(77.04) THA(total hip arthroplasty):全髋关节置换术;SLE(systemic lupus erythematosus):系统性红斑狼疮;BMI(body mass index):体质量指数;ASA(American Society of Anesthesiologists):美国麻醉医师协会;SLEDAI (SLE disease activity index):SLE疾病活动度评分 2.2 倾向性评分后SLE组与对照组基线特征比较
根据纳入和排除标准,对照组共入选862例患者,经倾向性评分进行1∶1匹配后,最终SLE组和对照组各163例患者纳入分析。基本特征方面,SLE组患者病程为9.8(4.0, 14.0)年,SLEDAI评分为0.19(0, 0),围术期糖皮质激素使用率为92.64%,BMI指标显著低于对照组(P<0.001)。在内科合并症方面,SLE组骨质疏松、呼吸系统、消化系统、泌尿系统、血液系统和继发或伴发其他风湿性疾病伴发率明显高于对照组(P均<0.05),其中呼吸系统受累主要为间质性肺病、肺动脉高压、肺血栓栓塞症,泌尿系统受累主要为狼疮性肾炎,风湿性疾病主要包括干燥综合征和抗磷脂抗体综合征。实验室检测显示,SLE组的术前血小板计数、淋巴细胞计数、血红蛋白水平、红细胞压积、白蛋白水平、血糖水平和活化部分凝血活酶时间均显著低于对照组,而C反应蛋白、红细胞沉降率和D-二聚体水平显著高于对照组(P均<0.05)。在手术相关指标方面,SLE组的ASA分级较高,手术时间较短,术中出血量较少(P均<0.05),详见表 2。
表 2 倾向性评分匹配后SLE组和对照组基线特征比较Table 2. Baseline data of the SLE group and the control group after propensity score matching指标 SLE组(n=163) 对照组(n=163) P值 BMI(x±s, kg/m2) 22.9±3.4 24.8±3.7 <0.001 ASA分级[n(%)] <0.001 ≤Ⅱ级 137(84.05) 158(96.93) >Ⅱ级 26(15.95) 5(3.07) 手术时间[M(P25, P75), min] 125.2(85.0, 175.0) 137.4(95.0, 175.0) 0.042 术中出血[M(P25, P75), mL] 324.7(200.0, 500.0) 421.8(200.0, 500.0) 0.005 合并症[n(%)] 骨质疏松 42(25.77) 15(9.20) <0.001 高血压 29(17.79) 26(15.95) 0.657 冠心病 3(1.84) 0 0.248 糖尿病 2(1.23) 3(1.84) >0.999 脑血管疾病 1(0.61) 1(0.61) >0.999 呼吸系统疾病 16(9.82) 2(1.23) 0.002 消化系统疾病 11(6.75) 1(0.61) 0.008 泌尿系统疾病 41(25.15) 0 <0.001 血液系统疾病 4(2.45) 0 0.018 继发或伴发其他风湿性疾病 17(10.43) 0 <0.001 术前实验室检测 血小板(x±s, ×109/L) 211.4±71.9 242.4±60.1 <0.001 淋巴细胞计数[M(P25, P75), ×109/L] 2.2(1.0, 1.8) 2.1(1.6, 2.3) <0.001 中性粒细胞计数[M(P25, P75), ×109/L] 4.6(2.7, 4.1) 3.7(2.6, 4.2) 0.843 血红蛋白(x±s, g/L) 124.3±13.6 132.8±13.3 <0.001 红细胞压积(x±s) 37.8±7.9 39.4±5.8 0.042 白蛋白(x±s, g/L) 39.6±4.0 42.1±3.5 <0.001 谷丙转氨酶[M(P25, P75), U/L] 20.1(11.0, 23.0) 20.6(12.0, 21.8) 0.907 肌酐[M(P25, P75), μmol/L] 69.7(51.0, 71.0) 61.1(49.3, 69.0) 0.340 血糖[M(P25, P75), mmol/L] 4.6(4.1, 4.7) 5.2(4.5, 5.6) <0.001 C反应蛋白[M(P25, P75), mg/L] 7.7(1.2, 6.2) 4.5(0.5, 2.3) <0.001 红细胞沉降率[M(P25, P75), mm/h] 25.1(13.0, 29.0) 12.5(8.0, 15.0) <0.001 活化部分凝血活酶时间[M(P25, P75), s] 27.8(25.0, 29.2) 28.6(26.5, 30.5) 0.004 D-二聚体[M(P25, P75), mg/L] 0.8(0.2, 0.7) 0.7(0.2, 0.5) <0.001 SLE、BMI、ASA: 同表 1 2.3 SLE组和对照组主要并发症及同种异体红细胞输注情况
术后并发症方面,SLE组术后30 d内主要急性并发症发生率显著高于对照组(8.59%比1.23%,P=0.005),相对危险度为1.081(95% CI: 1.028~1.136)。SLE组中,14例(8.59%)患者在术后30 d内出现主要急性并发症,包括1例(0.61%)上呼吸道感染;2例(1.23%)肺部感染,其中1例(0.61%)伴感染性休克;2例(1.23%)泌尿系感染;2例(1.23%)其他系统感染;1例(0.61%)应激性溃疡伴出血;2例(1.23%)切口愈合不良;1例(0.61%)切口感染;1例(0.61%)因输注血小板出现过敏性休克;1例(0.61%) SLE病情活动。相比之下,对照组中仅2例(1.23%)患者出现主要并发症,包括1例(0.61%)泌尿系感染和1例(0.61%)假体脱位。所有患者经积极保守治疗后均好转出院。此外,SLE组同种异体红细胞输注率与对照组差异无统计学意义(25.77%比17.18%,P=0.059)。详见表 3。
表 3 倾向性评分匹配后SLE组和对照组主要并发症及输血情况[n(%)]Table 3. Main complications and blood transfusion circumstances in the SLE group and the control group after propensity score matching [n(%)]指标 SLE组(n=163) 对照组(n=163) P值 主要并发症 14(8.59) 2(1.23) 0.005 上呼吸道感染 1(0.61) 0(0) 肺部感染 2(1.23) 0(0) 泌尿系感染 2(1.23) 1(0.61) 其他系统感染 2(1.23) 0(0) 消化系统 1(0.61) 0(0) 休克 2(1.23) 0(0) SLE活动 1(0.61) 0(0) 切口愈合不良 2(1.23) 0(0) 切口感染 1(0.61) 0(0) 假体脱位 0(0) 1(0.61) 异体红细胞输注率 42(25.77) 28(17.18) 0.059 SLE: 同表 1 3. 讨论
本研究分析了270例行THA术的SLE患者临床资料,发现患者术前SLEDAI评分为0.16分,围术期糖皮质激素使用率为77.04%,说明行THA术的SLE患者术前病情相对稳定。采用倾向性评分匹配法,分析SLE患者在THA后30 d内主要急性并发症的发生率及其影响因素,结果显示,在匹配前,SLE组并发症发生率为6.67%,61例(22.59%)接受同种异体红细胞输注。在性别、年龄、手术侧别匹配后,内科合并症方面,与对照组相比,SLE组在骨质疏松、呼吸系统、消化系统、泌尿系统、血液系统和继发或伴发其他风湿性疾病上存在显著差异,这符合SLE多器官受累的特征[16]。术前实验室检测显示,SLE组的血小板计数、淋巴细胞计数、血红蛋白、红细胞压积、白蛋白、血糖和活化部分凝血酶时间指标均低于对照组,而C反应蛋白、红细胞沉降率和D-二聚体水平较高,提示SLE组患者存在血液系统抑制、营养不良、炎症和高凝状态。手术相关指标方面,SLE组的ASA分级较高,但手术时间短,术中出血量少。并发症方面,SLE组术后30 d内主要急性并发症发生率显著高于对照组(8.59%比1.23%,P=0.005),差异具有统计学意义,相对危险度为1.081(95% CI: 1.028~1.136),而同种异体红细胞输注率与对照组无显著差异(25.8%比17.2%,P=0.059)。
近年大量研究显示,SLE患者由于其独特的病理生理特性,包括免疫功能紊乱、多系统受累以及长期激素和/或免疫抑制剂的使用,THA术后并发症的发生风险显著增加[1-3, 5, 9, 16],如各系统并发症、感染、切口不良事件及同种异体输血等。本研究中,SLE组主要并发症以感染和切口不良事件为主,与既往报道一致。此外,本研究还发现,SLE患者的术前红细胞压积、血红蛋白及白蛋白水平均显著低于对照组,提示其营养状态较差,这可能进一步增加术后感染的风险。切口不良事件在SLE患者中较为常见,包括切口延迟愈合和切口感染,可能与SLE患者长期使用激素导致的皮肤脆性及血管炎症相关[3, 13]。本研究显示,SLE组切口愈合不良发生率为1.85%,低于既往报道的4.30%~8.88%[1, 17]。值得注意的是,SLE病情活动是SLE患者接受THA的特有并发症。手术应激、药物控制不佳、感染等因素可能导致SLE复发,本研究中表现为狼疮性肾炎加重和补体水平降低,但发生率显著低于既往报道的2.27%[18]。这些差异可能得益于严格的围术期管理,如合理调整糖皮质激素用量、停用部分免疫抑制剂[9, 19]以及术中应用第二代头孢菌素预防感染。
SLE患者因病理性贫血和术中失血导致的输血需求是围术期关注的重点[2-3, 6, 16]。研究认为,SLE贫血主要与狼疮累及血液系统及肾功能不全导致促红细胞生成素生成减少有关[20]。此外,SLE患者可能由于血小板功能下降及抗凝血因子抗体的存在,因在围术期出血而引发贫血[1, 21]。然而,本研究结果显示,尽管SLE组术前血红蛋白水平显著低于对照组,但两组术后输血率差异并无统计学意义。这可能与术中总体出血量较少,以及合理使用氨甲环酸和血液回收系统等有效措施有关。
本研究存在一定的局限性。首先,本研究为回顾性研究,样本量相对较小,可能存在选择偏倚。其次,未对SLE及其他合并症的活动情况及疾病分期情况作分层分析,影响了结果的全面性和准确性。最后,本研究匹配的SLE病例中,SLEDAI最高仅为4分,无法代表病情控制不佳的SLE患者,可能低估了SLE病情活动患者的术后风险。
综上,本研究显示SLE患者行THA术后30 d内并发症发生率为8.59%,显著高于非SLE患者,而同种异体红细胞输注率相当。为保证SLE患者THA手术安全性,应尽可能在术前将患者病情控制稳定,并遵循指南进行严格的围术期管理。
作者贡献:刘润竹负责查阅文献、撰写论文;龙笑负责论文修订。利益冲突:所有作者均声明不存在利益冲突 -
图 1 面容测量学发展时间线总结
注:黄色为直接测量方法相关事件,蓝色为2D测量法相关事件,红色为3D测量法相关事件,紫色为人工智能与面容测量学结合相关事件
Figure 1. Timeline summary of the development of facial metrology
Note: Yellow for events related to direct measurement; Blue for events related to 2D measurement; Red for events related to 3D measurement; Purple for events related to the combination of AI and facial measurements
表 1 三种3D体表成像设备的优劣势比较
Table 1 Comparison of advantages and disadvantages of three 3D imaging devices
设备类型 原理 优势 缺点 代表设备 误差 激光扫描 激光束投射至面容表面,产生干涉条纹 成像速度快、应用范围广、对人体无害 准确性较低,成本高、体积大 Minolta Vivid 较大 结构光扫描 结构光栅条纹投射至面容表面,基于光学三角测量 操作方便、可捕捉运动对象、费用低、便携 面容还原度较低,测量精度低 FaceSCAN 1 mm以内 立体摄影 双目视觉原理,从多个角度捕捉面部特征并进行拼接 还原软组织表面颜色、纹理,成像速度快、环境要求低、数据存储方便 受被测量者面部运动的影响,需进行校准,价格昂贵 3dMDface、Vectra H 1 mm以内 表 2 人工智能在面容评估中的应用总结
Table 2 Summary of artificial intelligence applications in facial assessment
年份 第一作者 技术名称 识别对象 实现方法 基于图像类型 识别准确性 2018 Song[70] GES(AGing pattErn Subspace) 特纳综合征儿童面容 PCA+SVM 2D 84.6% 2019 Gurovich[66] DeepGestalt 遗传综合征及Noonan综合征亚型 计算机视觉和深度学习 2D 91.0% 2020 Bannister[79] NICP(Non-rigid Iterative Closest Point) 健康成年受试者面容标记点 基于成熟2D图像识别算法的3D投射 3D 平均误差2.5 mm 2021 Pan[74] - 特纳综合征儿童面容 DCNN 2D 97.3% 2022 Fu[81] - 胎儿酒精综合征 基于面部标记点的深度学习 3D 误差为0.59~1.58像素 2022 Hsieh[84] GestaltMatcher 罕见疾病 有监督DCCN 2D 78.1% 2022 Bannister[85] - 多种遗传综合征 贝叶斯分类器 3D 71% 2022 Rouxel[86] DeepGestalt 歌舞伎综合征及亚型 DCCN 2D 85.0% 2022 Hartmann[77] - 面部表情 ANN 3D 81.2% 2023 Bannister[19] - 多种遗传综合征 2D: CNN
3D: PCA2D+3D 2D: 68.0%
3D: 81.2%2023 Jiang[83] PointNet++ 半侧颜面萎缩患者面部标记点 CNN 3D 96.0% 2024 Chong[82] UNET 健康、肢端肥大、局限性硬皮病面容标记点 CNN 3D 平均误差1.4 mm 2024 Yang[87] 高分辨网络关键点检测 唇周标记点 深度学习 3D 平均误差1.3 mm PCA(principal component analysis):主成分分析;SVM(support vector machine):支持向量机;DCNN(deep convolutional neural networks):深度卷积神经网络;ANN(artificial neural networks),人工神经网络;CNN(convolutional neural networks):卷积神经网络;-: 未提及 表 3 面容测量学方法比较
Table 3 Comparison of different facial measuring methods
年份 第一作者 研究对象 2D测量 3D测量 直接测量 人工智能 结果 1995 Aung[34] 30名受试者面部43个标记点 - *(激光扫描) * - 唇周、鼻周误差<1 mm 1996 Gross[89] 不同表情下面部运动幅度,4名受试者面部15个标记点 * * - - 二维分析可能不足以评估最大动画期间的面部运动,三维分析可能更适合检测面部功能的临床差异 2006 Kovacs[39] 5名受试者面部48个标记点 - *(激光扫描) * - 误差超过2 mm 2006 Goos[90] 2名受试者面部7个标记点之间的距离 * *(激光扫描) - - 两种方法的差异较大 2007 Ghoddousi[17] 6名受试者面部14个标记点 * *(立体摄影) * - 3D测量法、直接测量法的重复性更好,2D测量结果变异性大,3D测量结果偏大 2011 Aynechi[31] 10名受试者面部18个线性指标 - *(立体摄影) * - 3D测量系统的精度较高,误差<2 mm 2013 Al-Anezi[40] 32名受试者面部23个标记点 - *(结构光扫描) - *(software) 平均误差<0.55 mm 2014 Kook[16] 12个模型的面部15个标记点 - *(立体摄影) * - 测量误差<0.7 mm,测量准确性、重复性均可 2014 Berssenbrügge[91] 50名受试者面部对称性指数 * *(立体摄影) - - 2D测量、3D测量下的对称性指数互相存在关联 2016 Ye[42] 10名受试者面部标记点 - *(立体摄影与结构光扫描) * - 两种3D测量法均可准确还原真实测量结果,其中立体摄影的可靠性更高 2016 Dindarogğlu[18] 80名受试者10个面部参数 * *(立体摄影) * - 3D测量法是最准确可靠的方法 2018 Gibelli[37] 50名受试者及人面部模型的面部标记点 - *(激光扫描与立体摄影) - - 激光扫描仪在模型扫描中可靠,在受试者中不可靠 2023 Gašparović[92] 39名受试者面部线性指标,术后自动取点 - *(立体摄影) - *(ICP) 自动取点测量的可重复性较高,与真实差距小 2023 Jiang[83] 135名受试者面部32个标记点 - *(立体摄影) - *(PointNet++,卷积神经网络) 平均误差1.46 mm,较准确 2023 Pellitteri[46] 30名受试者面部线性指标及分区 - *(立体摄影、结构光、智能手机应用程序) - - 三种方法的结果无明显差异 2024 Chong[82] 100名受试者面部20个标记点 - *(立体摄影) - *(CNN-UNET) 健康人群中平均误差为1.4 mm,疾病人群中误差增大 2024 Yang[87] 120名受试者唇周标记点 - *(立体摄影) - *(深度学习) 线性测量的平均误差为0.87 mm,角度测量的平均误差为5.62° *: 参与比较项目;-:该项未涉及 -
[1] Houle D, Govindaraju D R, Omholt S. Phenomics: the next challenge[J]. Nat Rev Genet, 2010, 11(12): 855-866. DOI: 10.1038/nrg2897
[2] Baynam G, Bauskis A, Pachter N, et al. 3-Dimensional facial analysis-facing precision public health[J]. Front Public Health, 2017, 5: 31.
[3] Hammond P, Hannes F, Suttie M, et al. Fine-grained facial phenotype-genotype analysis in Wolf-Hirschhorn syndrome[J]. Eur J Hum Genet, 2012, 20(1): 33-40. DOI: 10.1038/ejhg.2011.135
[4] Farkas L G, Posnick J C, Hreczko T M. Anthropometric growth study of the head[J]. Cleft Palate Craniofac J, 1992, 29(4): 303-308. DOI: 10.1597/1545-1569_1992_029_0303_agsoth_2.3.co_2
[5] Farkas L G, Katic M J, Forrest C R, et al. International anthropometric study of facial morphology in various ethnic groups/races[J]. J Craniofac Surg, 2005, 16(4): 615-646. DOI: 10.1097/01.scs.0000171847.58031.9e
[6] Farkas L G, Hajnis K. Faciometer[J]. Panminerva Med, 1969, 11(4): 165-169.
[7] Venkatadri G, Farkas L G, Kooiman J. Multipurpose anthropometric facial anglemeter[J]. Plast Reconstr Surg, 1992, 90(3): 507-510. DOI: 10.1097/00006534-199209000-00027
[8] Farkas L G, Feiglová B, Klásková O. Several data of the analysis of 1, 000 etiopathological questionnaires in patients with cleft lips and palates[J]. Panminerva Med, 1969, 11(1): 59-63.
[9] Farkas L G, Munro I R, Kolar J C. Abnormal measurements and disproportions in the face of Down's syndrome patients: preliminary report of an anthropometric study[J]. Plast Reconstr Surg, 1985, 75(2): 159-169. DOI: 10.1097/00006534-198502000-00002
[10] Zhuang Z Q, Landsittel D, Benson S, et al. Facial anthropometric differences among gender, ethnicity, and age groups[J]. Ann Occup Hyg, 2010, 54(4): 391-402.
[11] Du L L, Zhuang Z Q, Guan H Y, et al. Head-and-face anthropometric survey of Chinese workers[J]. Ann Occup Hyg, 2008, 52(8): 773-782.
[12] DiBernardo B E, Adams R L, Krause J, et al. Photographic standards in plastic surgery[J]. Plast Reconstr Surg, 1998, 102(2): 559-568. DOI: 10.1097/00006534-199808000-00045
[13] Yavuzer R, Smirnes S, Jackson I T. Guidelines for standard photography in plastic surgery[J]. Ann Plast Surg, 2001, 46(3): 293-300. DOI: 10.1097/00000637-200103000-00016
[14] Lin C S, Shaari R, Alam M K, et al. Photogrammetric analysis of nasolabial angle and mentolabial angle norm in Malaysian adults[J]. Bangladesh J Med Sci, 2013, 12(2): 209-219. DOI: 10.3329/bjms.v12i2.14951
[15] Packiriswamy V, Kumar P, Rao K G M. Photogrammetric analysis of palpebral fissure dimensions and its position in Malaysian South Indian ethnic adults by gender[J]. N Am J Med Sci, 2012, 4(10): 458-462. DOI: 10.4103/1947-2714.101984
[16] Kook M S, Jung S, Park H J, et al. A comparison study of different facial soft tissue analysis methods[J]. J Craniomaxillofac Surg, 2014, 42(5): 648-656. DOI: 10.1016/j.jcms.2013.09.010
[17] Ghoddousi H, Edler R, Haers P, et al. Comparison of three methods of facial measurement[J]. Int J Oral Maxillofac Surg, 2007, 36(3): 250-258. DOI: 10.1016/j.ijom.2006.10.001
[18] Dindaroğlu F, Kutlu P, Duran G S, et al. Accuracy and reliability of 3D stereophotogrammetry: a comparison to direct anthropometry and 2D photogrammetry[J]. Angle Orthod, 2016, 86(3): 487-494. DOI: 10.2319/041415-244.1
[19] Bannister J J, Wilms M, Aponte J D, et al. Comparing 2D and 3D representations for face-based genetic syndrome diagnosis[J]. Eur J Hum Genet, 2023, 31(9): 1010-1016. DOI: 10.1038/s41431-023-01308-w
[20] Van Vlijmen O J C, Maal T, Bergé S J, et al. A comparison between 2D and 3D cephalometry on CBCT scans of human skulls[J]. Int J Oral Maxillofac Surg, 2010, 39(2): 156-160. DOI: 10.1016/j.ijom.2009.11.017
[21] Anas I Y, Bamgbose B O, Nuhu S. A comparison between 2D and 3D methods of quantifying facial morphology[J]. Heliyon, 2019, 5(6): e01880. DOI: 10.1016/j.heliyon.2019.e01880
[22] Sforza C, Laino A, D'Alessio R, et al. Soft-tissue facial characteristics of attractive and normal adolescent boys and girls[J]. Angle Orthod, 2008, 78(5): 799-807. DOI: 10.2319/091207-431.1
[23] Wang D, Qian G, Zhang M, et al. Differences in horizontal, neoclassical facial canons in Chinese (Han) and North American Caucasian populations[J]. Aesthetic Plast Surg, 1997, 21(4): 265-269. DOI: 10.1007/s002669900123
[24] De Coster P, De Pauw G, Martens L, et al. Craniofacial structure in Marfan syndrome: a cephalometric study[J]. Am J Med Genet A, 2004, 131A(3): 240-248. DOI: 10.1002/ajmg.a.30393
[25] Ting B L, Mathur D, Loeys B L, et al. The diagnostic value of the facial features of Marfan syndrome[J]. J Child Orthop, 2010, 4(6): 545-551. DOI: 10.1007/s11832-010-0295-1
[26] Sutton V R, Hopkins B J, Eble T N, et al. Facial and physical features of Aicardi syndrome: infants to teenagers[J]. Am J Med Genet A, 2005, 138A(3): 254-258. DOI: 10.1002/ajmg.a.30963
[27] Maclean K, Smith J, St Heaps L, et al. Axenfeld-Rieger malformation and distinctive facial features: clues to a recognizable 6p25 microdeletion syndrome[J]. Am J Med Genet A, 2005, 132A(4): 381-385. DOI: 10.1002/ajmg.a.30274
[28] Moore E S, Ward R E, Wetherill L F, et al. Unique facial features distinguish fetal alcohol syndrome patients and controls in diverse ethnic populations[J]. Alcohol Clin Exp Res, 2007, 31(10): 1707-1713. DOI: 10.1111/j.1530-0277.2007.00472.x
[29] Cai X X, Chen Y, Li Q, et al. Anthropometric analysis on the ocular region morphology of children and young adults in Chinese Han population[J]. Ophthalmic Plast Reconstr Surg, 2019, 35(4): 326-332. DOI: 10.1097/IOP.0000000000001245
[30] Wen Y F, Wong H M, McGrath C P. A longitudinal study of facial growth of Southern Chinese in Hong Kong: comprehensive photogrammetric analyses[J]. PLoS One, 2017, 12(10): e0186598. DOI: 10.1371/journal.pone.0186598
[31] Aynechi N, Larson B E, Leon-Salazar V, et al. Accuracy and precision of a 3D anthropometric facial analysis with and without landmark labeling before image acquisition[J]. Angle Orthod, 2011, 81(2): 245-252. DOI: 10.2319/041810-210.1
[32] Sforza C, Dellavia C, Tartaglia G M, et al. Morphometry of the ear in Down's syndrome subjects: a three-dimensional computerized assessment[J]. Int J Oral Maxillofac Surg, 2005, 34(5): 480-486. DOI: 10.1016/j.ijom.2004.10.005
[33] Bhatia G, Vannier M W, Smith K E, et al. Quantification of facial surface change using a structured light scanner[J]. Plast Reconstr Surg, 1994, 94(6): 768-774. DOI: 10.1097/00006534-199411000-00003
[34] Aung S C, Ngim R C K, Lee S T. Evaluation of the laser scanner as a surface measuring tool and its accuracy compared with direct facial anthropometric measurements[J]. Br J Plast Surg, 1995, 48(8): 551-558. DOI: 10.1016/0007-1226(95)90043-8
[35] Ras F, Habets L L, Van Ginkel F C, et al. Quantification of facial morphology using stereophotogrammetry--demonstration of a new concept[J]. J Dent, 1996, 24(5): 369-374. DOI: 10.1016/0300-5712(95)00081-X
[36] Moss J P, Linney A D, Grindrod S R, et al. A laser scanning system for the measurement of facial surface morphology[J]. Opt Lasers Eng, 1989, 10(3/4): 179-190.
[37] Gibelli D, Pucciarelli V, Caplova Z, et al. Validation of a low-cost laser scanner device for the assessment of three-dimensional facial anatomy in living subjects[J]. J Craniomaxillofac Surg, 2018, 46(9): 1493-1499. DOI: 10.1016/j.jcms.2018.06.009
[38] Cao R K, Chen B B, Xu H, et al. Accuracy of three-dimensional optical devices for facial soft-tissue measurement in clinical practice of stomatology: a PRISMA systematic review[J]. Medicine (Baltimore), 2022, 101(47): e31922. DOI: 10.1097/MD.0000000000031922
[39] Kovacs L, Zimmermann A, Brockmann G, et al. Three-dimensional recording of the human face with a 3D laser scanner[J]. J Plast Reconstr Aesthet Surg, 2006, 59(11): 1193-1202. DOI: 10.1016/j.bjps.2005.10.025
[40] Al-Anezi T, Khambay B, Peng M J, et al. A new method for automatic tracking of facial landmarks in 3D motion captured images (4D)[J]. Int J Oral Maxillofac Surg, 2013, 42(1): 9-18. DOI: 10.1016/j.ijom.2012.10.035
[41] Ma L L, Xu T M, Lin J X. Validation of a three-dimensional facial scanning system based on structured light techniques[J]. Comput Methods Programs Biomed, 2009, 94(3): 290-298. DOI: 10.1016/j.cmpb.2009.01.010
[42] Ye H Q, Lv L W, Liu Y S, et al. Evaluation of the accuracy, reliability, and reproducibility of two different 3D face-scanning systems[J]. Int J Prosthodont, 2016, 29(3): 213-218. DOI: 10.11607/ijp.4397
[43] Hajeer M Y, Millett D T, Ayoub A F, et al. Applications of 3D imaging in orthodontics: part Ⅰ[J]. J Orthod, 2004, 31(1): 62-70. DOI: 10.1179/146531204225011346
[44] Gibelli D, Pucciarelli V, Cappella A, et al. Are portable stereophotogrammetric devices reliable in facial imaging? A validation study of VECTRA H1 device[J]. J Oral Maxillofac Surg, 2018, 76(8): 1772-1784. DOI: 10.1016/j.joms.2018.01.021
[45] Camison L, Bykowski M, Lee W W, et al. Validation of the Vectra H1 portable three-dimensional photogrammetry system for facial imaging[J]. Int J Oral Maxillofac Surg, 2018, 47(3): 403-410. DOI: 10.1016/j.ijom.2017.08.008
[46] Pellitteri F, Scisciola F, Cremonini F, et al. Accuracy of 3D facial scans: a comparison of three different scanning system in an in vivo study[J]. Prog Orthod, 2023, 24(1): 44. DOI: 10.1186/s40510-023-00496-x
[47] Chong Y M, Liu X Y, Shi M, et al. Three-dimensional facial scanner in the hands of patients: validation of a novel application on iPad/iPhone for three-dimensional imaging[J]. Ann Transl Med, 2021, 9(14): 1115. DOI: 10.21037/atm-21-1620
[48] Yi C R, Kim M J, Kim S C, et al. Comparison of facial proportions between eastern and western attractive young women[J]. J Craniofac Surg, 2020, 31(3): e303-e306. DOI: 10.1097/SCS.0000000000006196
[49] Kim Y C, Kwon J G, Kim S C, et al. Comparison of periorbital anthropometry between beauty pageant contestants and ordinary young women with Korean ethnicity: a three-dimensional photogrammetric analysis[J]. Aesthetic Plast Surg, 2018, 42(2): 479-490. DOI: 10.1007/s00266-017-1040-7
[50] Chong Y M, Li J Y, Liu X Y, et al. Three-dimensional anthropometric analysis of eyelid aging among Chinese women[J]. J Plast Reconstr Aesthet Surg, 2021, 74(1): 135-142. DOI: 10.1016/j.bjps.2020.08.010
[51] Modabber A, Peters F, Galster H, et al. Three-dimensional evaluation of important surgical landmarks of the face during aging[J]. Ann Anat, 2020, 228: 151435. DOI: 10.1016/j.aanat.2019.151435
[52] Lambros V. Facial aging: a 54-year, three-dimensional population study[J]. Plast Reconstr Surg, 2020, 145(4): 921-928. DOI: 10.1097/PRS.0000000000006711
[53] Kim B R, Kim M, Kim J W, et al. Efficacy and safety of the micro-insulated needle radiofrequency device for reduction of submental fat[J]. Dermatol Surg, 2023, 49(4): 389-394. DOI: 10.1097/DSS.0000000000003723
[54] Nikolis A, Bertucci V, Solish N, et al. An objective, quantitative assessment of flexible hyaluronic acid fillers in lip and perioral enhancement[J]. Dermatol Surg, 2021, 47(5): e168-e173. DOI: 10.1097/DSS.0000000000002917
[55] Mailey B, Baker J L, Hosseini A, et al. Evaluation of facial volume changes after rejuvenation surgery using a 3-dimensional camera[J]. Aesthet Surg J, 2016, 36(4): 379-387. DOI: 10.1093/asj/sjv226
[56] Dolci C, Sansone V A, Gibelli D, et al. Distinctive facial features in Andersen-Tawil syndrome: a three-dimensional stereophotogrammetric analysis[J]. Am J Med Genet A, 2021, 185(3): 781-789. DOI: 10.1002/ajmg.a.62040
[57] Jayaratne Y S N, Elsharkawi I, Macklin E A, et al. The facial morphology in Down syndrome: a 3D comparison of patients with and without obstructive sleep apnea[J]. Am J Med Genet A, 2017, 173(11): 3013-3021. DOI: 10.1002/ajmg.a.38399
[58] Ji C, Yao D, Li M Y, et al. A study on facial features of children with Williams syndrome in China based on three-dimensional anthropometric measurement technology[J]. Am J Med Genet A, 2020, 182(9): 2102-2109. DOI: 10.1002/ajmg.a.61750
[59] Lin H C, Le Hoang P, Hutchinson A, et al. Alagille syndrome in a Vietnamese cohort: mutation analysis and assessment of facial features[J]. Am J Med Genet A, 2012, 158A(5): 1005-1013. DOI: 10.1002/ajmg.a.35255
[60] Guo X P, Meng T, Huang J Z, et al. 3D facial analysis in acromegaly: gender-specific features and clinical correlations[J]. Front Endocrinol (Lausanne), 2018, 9: 722. DOI: 10.3389/fendo.2018.00722
[61] Wang M H, Lin J D, Chang C N, et al. Acromegaly determination using discriminant analysis of the three-dimensional facial classification in Taiwanese[J]. Neuro Endocrinol Lett, 2017, 38(4): 301-309.
[62] Du F Z, Chen Q, Wang X J, et al. Long-term facial changes and clinical correlations in patients with treated acromegaly: a cohort study[J]. Eur J Endocrinol, 2021, 184(2): 231-241. DOI: 10.1530/EJE-20-0962
[63] Botticelli S, Pedersen T K, Küseler A, et al. Novel 3-D analysis for the assessment of cleft dimensions on digital models of infants with unilateral cleft lip and palate[J]. Cleft Palate Craniofac J, 2019, 56(1): 127-133. DOI: 10.1177/1055665618770795
[64] Desmedt D J, Maal T J, Kuijpers M A, et al. Nasolabial symmetry and esthetics in cleft lip and palate: analysis of 3D facial images[J]. Clin Oral Investig, 2015, 19(8): 1833-1842. DOI: 10.1007/s00784-015-1445-0
[65] Ganske I M, Langa O C, Mulliken J B. Philtral ridge projection in repaired unilateral cleft lip: Three-Dimensional anthropometry[J]. Plast Reconstr Surg, 2021, 147(1): 141-147. DOI: 10.1097/PRS.0000000000007460
[66] Gurovich Y, Hanani Y, Bar O, et al. Identifying facial phenotypes of genetic disorders using deep learning[J]. Nat Med, 2019, 25(1): 60-64. DOI: 10.1038/s41591-018-0279-0
[67] Cornejo J Y R, Pedrini H, Machado-Lima A, et al. Down syndrome detection based on facial features using a geometric descriptor[J]. J Med Imaging (Bellingham), 2017, 4(4): 044008.
[68] Zhao Q, Rosenbaum K, Okada K, et al. Automated down syndrome detection using facial photographs[J]. Annu Int Conf IEEE Eng Med Biol Soc, 2013, 2013: 3670-3673.
[69] Dhanamjayulu C, Nizhal U N, Maddikunta P K R, et al. Identification of malnutrition and prediction of BMI from facial images using real-time image processing and machine learning[J]. IET Image Process, 2022, 16(3): 647-658. DOI: 10.1049/ipr2.12222
[70] Song W A, Lei Y, Chen S, et al. Multiple facial image features-based recognition for the automatic diagnosis of turner syndrome[J]. Comput Ind, 2018, 100: 85-95. DOI: 10.1016/j.compind.2018.03.021
[71] Meng T, Guo X P, Lian W, et al. Identifying facial features and predicting patients of acromegaly using three-dimensional imaging techniques and machine learning[J]. Front Endocrinol (Lausanne), 2020, 11: 492. DOI: 10.3389/fendo.2020.00492
[72] Guarin D L, Yunusova Y, Taati B, et al. Toward an automatic system for computer-aided assessment in facial palsy[J]. Facial Plast Surg Aesthet Med, 2020, 22(1): 42-49. DOI: 10.1089/fpsam.2019.29000.gua
[73] Guarin D L, Dusseldorp J, Hadlock T A, et al. A machine learning approach for automated facial measurements in facial palsy[J]. JAMA Facial Plast Surg, 2018, 20(4): 335-337. DOI: 10.1001/jamafacial.2018.0030
[74] Pan Z X, Shen Z, Zhu H J, et al. Clinical application of an automatic facial recognition system based on deep learning for diagnosis of Turner syndrome[J]. Endocrine, 2021, 72(3): 865-873. DOI: 10.1007/s12020-020-02539-3
[75] Latorre-Pellicer A, Ascaso Á, Trujillano L, et al. Evaluat-ing Face2Gene as a tool to identify Cornelia de Lange syndrome by facial phenotypes[J]. Int J Mol Sci, 2020, 21(3): 1042. DOI: 10.3390/ijms21031042
[76] Wilamowska K, Shapiro L, Heike C. Classification of 3D face shape in 22Q11.2 deletion syndrome[C]//Proceedings of the Sixth IEEE international conference on Symposium on Biomedical Imaging: From Nano to Macro. Piscataway, NJ: IEEE Press, 2009: 534-537.
[77] Hartmann T J, Hartmann J B J, Friebe-Hoffmann U, et al. Novel method for three-dimensional facial expression recognition using self-normalizing neural networks and mobile devices[J]. Geburtshilfe Frauenheilkd, 2022, 82(9): 955-969. DOI: 10.1055/a-1866-2943
[78] Yang Y Y, Zhang M Y, Jin L, et al. Gender- and age-related characterization of lip morphology: a three-dimensi-onal analysis in a Chinese population[J]. Aesthet Surg J, 2023, 43(12): NP990-NP1000. DOI: 10.1093/asj/sjad270
[79] Bannister J J, Crites S R, Aponte J D, et al. Fully automatic landmarking of syndromic 3D facial surface scans using 2D images[J]. Sensors (Basel), 2020, 20(11): 3171. DOI: 10.3390/s20113171
[80] Chong Y M, Xiao Y D, Li Z M, et al. Introduction and validation of an automatic and low-cost three-dimensional facial imaging system: a comparison to direct anthropometry and Vectra H1[J]. Quant Imaging Med Surg, 2023, 13(5): 2922-2932. DOI: 10.21037/qims-22-900
[81] Fu Z Y, Jiao J B, Suttie M, et al. Facial anatomical landmark detection using regularized transfer learning with application to fetal alcohol syndrome recognition[J]. IEEE J Biomed Health Inform, 2022, 26(4): 1591-1601. DOI: 10.1109/JBHI.2021.3110680
[82] Chong Y M, Du F Z, Ma X D, et al. Automated anatomical landmark detection on 3D facial images using U-NET-based deep learning algorithm[J]. Quant Imaging Med Surg, 2024, 14(3): 2466-2474. DOI: 10.21037/qims-22-1108
[83] Jiang C, Jiang F L, Xie Z K, et al. Evaluation of automated detection of head position on lateral cephalometric radiographs based on deep learning techniques[J]. Ann Anat, 2023, 250: 152114. DOI: 10.1016/j.aanat.2023.152114
[84] Hsieh T C, Bar-Haim A, Moosa S, et al. GestaltMatcher facilitates rare disease matching using facial phenotype descriptors[J]. Nat Gene, 2022, 54(3): 349-357. DOI: 10.1038/s41588-021-01010-x
[85] Bannister J J, Wilms M, Aponte J D, et al. A Deep Invertible 3D Facial Shape Model For Interpretable Genetic Syndrome Diagnosis[J]. IEEE J Biomed Health Inform, 2022, 26(7): 3229-3239. DOI: 10.1109/JBHI.2022.3164848
[86] Rouxel F, Yauy K, Boursier G, et al. Using deep-neural-network-driven facial recognition to identify distinct Kabuki syndrome 1 and 2 gestalt[J]. Eur J Hum Genet, 2022, 30(6): 682-686. DOI: 10.1038/s41431-021-00994-8
[87] Yang Y Y, Zhang M Y, An Y C, et al. Automated 3D perioral landmark detection using high-resolution network: artificial intelligence-based anthropometric analysis[J]. Aesthet Surg J, 2024, 44(8): NP606-NP612. DOI: 10.1093/asj/sjae103
[88] Meintjes E M, Douglas T S, Martinez F, et al. A stereo-photogrammetric method to measure the facial dysmorphology of children in the diagnosis of fetal alcohol syndrome[J]. Med Eng Phys, 2002, 24(10): 683-689. DOI: 10.1016/S1350-4533(02)00114-5
[89] Gross M M, Trotman C A, Moffatt K S. A comparison of three-dimensional and two-dimensional analyses of facial motion[J]. Angle Orthod, 1996, 66(3): 189-194.
[90] Goos M I M, Alberink I B, Ruifrok A C C. 2D/3D image (facial) comparison using camera matching[J]. Forensic Sci Int, 2006, 163(1/2): 10-17.
[91] Berssenbrügge P, Berlin N F, Kebeck G, et al. 2D and 3D analysis methods of facial asymmetry in comparison[J]. J Craniomaxillofac Surg, 2014, 42(6): e327-e334. DOI: 10.1016/j.jcms.2014.01.028
[92] Gašparović B, Morelato L, Lenac K, et al. Comparing direct measurements and three-dimensional (3D) scans for evaluating facial soft tissue[J]. Sensors (Basel), 2023, 23(5): 2412. DOI: 10.3390/s23052412