A machine learning approach to investigate the materials science of enamel aging

Understanding aging of tooth tissues is critical to the development of patient-centric oral healthcare. Yet, the traditional methods for analyzing the composition-structure-property relationships of hard tissues have limitations when considering aging and other factors.

To apply unsupervised machine learning tools to pursue an understanding of relationships between the composition and mechanical behavior of aging enamel.

Molar teeth were collected from primary (age ≤ 8), young adult (24 ≤ age ≤ 46) and old adult (55 ≤ age) donors. The hardness and elastic modulus were quantified using nanoindentation as a function of distance from the Dentin Enamel Junction (DEJ) within the cervical, cuspal and inter-cuspal regions of the enamel crown. Similarly, a co-located analysis of the chemical composition and structure was performed using Raman spectroscopy. A Self-Organizing Maps (SOMs) algorithm was implemented to identify multi-dimensional composition-property relationships.

The hardness and elastic modulus are positively correlated to crystallinity and negatively correlated with carbonate substitution. Furthermore, the effects from fluoridation on the age-dependent properties of enamel is non-linear and depends on its location. The contributions of fluoridation to the enamel properties are different in the cervical and non-cervical regions and appear to be unique within primary and senior adult teeth.

Based on the findings, unsupervised learning methods can reveal complicated non-linear structure-property relationships in tooth tissues and help to understand the materials science of aging and its consequences.