Classification Algorithms for Early Tooth Demineralization Assessment by Impedance Spectroscopy

Oral caries is one of the most common oral diseases worldwide, affecting about 2.4 billion people. This phenomenon always starts with enamel demineralization, eventually progressing to tooth cavitation and loss when not properly treated. Nowadays, the standard diagnostic techniques to detect demineralization strongly depend on the operator's expertise and are characterized by fairly low sensitivity and specificity, and/or involve ionizing radiation. This study investigates the feasibility of a non-invasive, effective, rapid, and radiation-free approach employing impedance spectroscopy for early caries detection. Two binary classifiers were developed for automated assessment and validated using a dataset obtained by in vitro demineralization of human teeth. A computationally efficient single-neuron classifier, utilizing a single impedance phase measurement at 15 Hz, achieved 88% accuracy, offering a lightweight, low-power solution suitable for microcontroller implementation and rapid measurements. A Multi-Layer Perceptron (MLP) classifier, utilizing equivalent circuit element values, yielded a similar accuracy of 86%. A prototype of a diagnostic portable tool was developed and characterized, demonstrating reliable impedance phase measurement (uncertainty < 2 degrees). The performance of these classifiers meets or exceeds the existing AI-based methodologies for caries detection relying on radiographic data. This work introduces a novel application of AI to tooth impedance spectra, addressing a significant research gap in non-invasive diagnostics and laying the foundation for a novel, accessible, and accurate tool for early caries management.