Back to search

National and Kapodistrian University of Athens

Ανάπτυξη μεθόδων υπολογιστικής γεωμετρίας για την ορθή ταξινόμηση αναμεμειγμένων ανθρώπινων σκελετικών καταλοίπων

Abstract

dc:description

The aim of this thesis was to explore the integration of computational geometry in osteological analysis and develop methodologies based on this approach for sorting commingled human skeletal remains. Additionally, subsequent analyses aimed to evaluate the developed methodologies on an archaeological and historical sample, serving as a reference study for their application to unknown skeletal remains of different populations and time periods. The developed algorithms were based on 3D models of long bones and utilized their diaphyseal cross-sectional geometric properties. The different research questions of this work concerned the estimation of biological sex from ulnae, the sorting of diaphyseal fragments and different long bones, as well as the application of the developed algorithms on a skeletal sample from a different chronological and geographical context. The dissertation is structured into six chapters: Chapter 1 provides a review of the existing literature regarding the significance of osteological analyses and current methodologies, Chapter 2 outlines the aims of the present dissertation and the gap in the literature it seeks to address, and Chapter 3 describes the skeletal material and the methodologies employed for the development and evaluation of the algorithms. The results are presented in Chapter 4, followed by their discussion and interpretation in light of the existing literature (Chapter 5) and the final conclusions of the dissertation (Chapter 6). The algorithms were trained on contemporary skeletal material from the Athens Collection, consisting of adult individuals with no signs of pathology, from individual burials. For the sexual dimorphism study, the sample consisted of 215 individuals (120 males, 95 females) of known biological sex and age-at-death, while the sorting methodologies used 396 femora, 422 tibiae, 415 humeri, and 406 ulnae bones. The archaeological and historical sample consisted of two subsets: an ancient Greek subset, ranging from the Early Neolithic to the Early Christian period, and a historical British subset, from the Early Christian period to the 19th century. In total, 95 individuals (46 males, 29 females, 10 of unknown biological sex) and 314 skeletal elements (104 femora, 129 tibiae, and 81 humeri) were examined. Where sample size permitted, the algorithms trained on the reference dataset were evaluated separately for the two populations. The developed methodologies employed Machine Learning algorithms for the classification process, such as Support Vector Machine models. Algorithm performance during the initial development and the application process was measured using evaluation metrics such as classification accuracy, true negative rate, and the percentage of false rejections of true skeletal matches. The results were encouraging in most analyses, outperforming previously published methods. The highest cross-validated accuracy achieved for sex estimation was 98.15%, using a combination of variables that displayed statistically significant sexual dimorphism and similar directional asymmetry, exceeding the accuracy of models based on traditional linear measurements. The development of a sorting methodology focused on long-bone diaphyseal fragments addressed a gap in literature regarding the association of bone fragments, yielding satisfactory results in distinguishing probable from improbable pairs, although overall sorting accuracy remained low. Similarly, the method for sorting different long bones, which involved the association of non-adjacent skeletal elements, yielded similar or better results than similar studies, while being fully automated. Despite the higher classification accuracies achieved in this scenario, the method is also recommended as an initial screening step. The evaluation on the archaeological and historical sample produced interesting results. The accuracy values for sex estimation and sorting of bilateral elements indicated that algorithms based on diaphyseal cross-sectional geometric properties can be used with relative confidence on samples from different populations and time periods. However, the scenarios involving fragment classification and sorting of different skeletal elements were less straightforward, with parameters such as sample size or the strictness of the initial screening range significantly influencing the outcome. The true negative rate values for mismatched pairs was high in most cases, but the high number of false negative instances does not allow the sole reliance on these algorithms. Instead, a combination of sorting methods is recommended when studying unknown samples. Lastly, no cross-population trends were observed regarding the performance of the methods; however, a larger sample size is required for more robust conclusions to be drawn. The utilized code and the developed algorithms, which have been incorporated in a graphical user interface as tools, are freely available in corresponding GitHub repositories, aiming for the facilitation of their use and further optimisation.

Degree

thesis:*
Grantor dc:publisher
National and Kapodistrian University of Athens
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Garoufi, Nefeli
  • Γαρούφη, Νεφέλη

Subjects

dc:subject × 28

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/61573

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Garoufi, Nefeli; Γαρούφη, Νεφέλη. Ανάπτυξη μεθόδων υπολογιστικής γεωμετρίας για την ορθή ταξινόμηση αναμεμειγμένων ανθρώπινων σκελετικών καταλοίπων. National and Kapodistrian University of Athens, 2026. http://hdl.handle.net/10442/hedi/61573