Use this URL to cite or link to this record in EThOS:
Title: Using 3D point cloud data and machine learning to assess skeletal remains
Author: Lam, Jessica F.
ISNI:       0000 0004 9358 9140
Awarding Body: University of Leicester
Current Institution: University of Leicester
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Access from Institution:
The purpose of this PhD project was to create a new method of sex assessment for crania using 3D point cloud data and machine learning. Through the process of investigating sexual dimorphism, this project has been the first to define the “discrimination factor”, which provides both researchers and practitioners of forensic anthropology a new tool for quantifying sexual dimorphism and comparing morphological traits. This project also created a ground-truth database of 3D point cloud data by using structured light scanning to document 534 crania (263 belonging to females and 271 belonging to males) from four diverse skeletal collections (located in the United Kingdom, Japan, Italy, and South Africa). A program called CraniAlign was created in conjunction with Clotho AI to process the 3D point cloud data in a manner that was transparent, reliable, and allowed for automation. In the first study of its kind, CraniAlign was compared to DAVID 4, which is the industry standard, in order to demonstrate that proprietary algorithms are not ideal for research. Finally, the 3D point cloud data of 316 individuals (134 female, 182 male) were used to train and test artificial neural networks. Three methods were successfully created – one that sought to classify individuals according to sex regardless of the population to which they belonged; one for classifying individuals according to sex and population; and one for classifying individuals into population groups regardless of sex. All three methods yielded training accuracies of 97.1% - 100.0% and evaluation accuracies of 87.5% - 92.5%. This project was therefore the first to apply deep learning to the problems of sex, population, and population-specific sex classification using the entire geometry of the cranium, and has successfully established three methods with unprecedented performances when tested on samples which were not involved in the training of the models.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
Keywords: Machine Learning ; 3D Point Cloud Data ; Skeletal Remains ; Archaeology ; Thesis