University of Pretoria, Department of Anatomy, Faculty of Health Sciences, Tswelopele Building, Private Bag X323, Prinshof 349-Jr, Pretoria 0084, South Africa.
Forensic Sci Int. 2024 Jun;359:112026. doi: 10.1016/j.forsciint.2024.112026. Epub 2024 Apr 16.
Forensic Facial Approximation (FFA) has evolved, with techniques advancing to refine the intercorrelation between the soft-tissue facial profile and the underlying skull. FFA has become essential for identifying unknown persons in South Africa, where the high number of migrant and illegal labourers and many unidentified remains make the identification process challenging. However, existing FFA methods are based on American or European standards, rendering them inapplicable in a South African context. We addressed this issue by conducting a study to create prediction models based on the relationships between facial morphology and known factors, such as population affinity, sex, and age, in white South African and French samples. We retrospectively collected 184 adult cone beam computed tomography (CBCT) scans representing 76 white South Africans (29 males and 47 females) and 108 French nationals (54 males and 54 females) to develop predictive statistical models using a projection onto latent structures regression algorithm (PLSR). On training and untrained datasets, the accuracy of the estimated soft-tissue shape of the ears, eyes, nose, and mouth was measured using metric deviations. The predictive models were optimized by integrating additional variables such as sex and age. Based on trained data, the prediction errors for the ears, eyes, nose, and mouth ranged between 1.6 mm and 4.1 mm for white South Africans; for the French group, they ranged between 1.9 mm and 4.2 mm. Prediction errors on non-trained data ranged between 1.6 mm and 4.3 mm for white South Africans, whereas prediction errors ranging between 1.8 mm and 4.3 mm were observed for the French. Ultimately, our study provided promising predictive models. Although the statistical models can be improved, the inherent variability among individuals restricts the accuracy of FFA. The predictive validity of the models was improved by including sex and age variables and considering population affinity. By integrating these factors, more customized and accurate predictive models can be developed, ultimately strengthening the effectiveness of forensic analysis in the South African region.
法医面型重构（FFA）技术不断发展，各种技术手段得以精进，以完善软组织面型与颅骨底层之间的相互关系。FFA 对于识别南非的身份不明人员至关重要，南非的移民和非法劳工人数众多，还有许多未被识别的遗骸，这使得识别过程颇具挑战。然而，现有的 FFA 方法基于美国或欧洲的标准，因此在南非的语境下并不适用。为了解决这个问题，我们进行了一项研究，旨在针对南非白人和法国人群，创建基于面部形态与已知因素（如种群亲缘关系、性别和年龄）之间关系的预测模型。我们回顾性地收集了 184 例成年锥形束 CT（CBCT）扫描图像，其中包括 76 名南非白人（29 名男性和 47 名女性）和 108 名法国国民（54 名男性和 54 名女性），以使用投影到潜在结构回归算法（PLSR）开发预测性统计模型。在训练和未训练数据集上，使用度量偏差测量估计的耳朵、眼睛、鼻子和嘴巴的软组织形状的准确性。通过整合性别和年龄等附加变量对预测模型进行了优化。基于训练数据，南非白人的耳朵、眼睛、鼻子和嘴巴的预测误差在 1.6mm 至 4.1mm 之间；对于法国人群，预测误差在 1.9mm 至 4.2mm 之间。对于未训练数据，南非白人的预测误差在 1.6mm 至 4.3mm 之间，而法国人群的预测误差在 1.8mm 至 4.3mm 之间。最终，我们的研究提供了有前景的预测模型。尽管统计模型可以得到改进，但个体之间的固有变异性限制了 FFA 的准确性。通过纳入性别和年龄变量并考虑种群亲缘关系，提高了模型的预测有效性。通过整合这些因素，可以开发出更具针对性和更准确的预测模型，最终加强法医分析在南非地区的有效性。
