An Explainable AI Approach for Periodontitis Severity Classification from Intraoral Images

Periodontitis is a prevalent inflammatory condition of the oral cavity, with severity ranging from moderate to severe. Early and precise classification is crucial for effective treatment planning and better patient outcomes. This study aims to extract textural, shape and statistical features from intraoral photographic images, which are routinely acquired in clinical practice. The novelty of this approach lies in a radiomics-inspired paradigm and machine learning for photographic imaging. A key strength of this study is the incorporation of explainable AI through partial dependence plots, ensuring validation of the proposed method and uncovering clinically meaningful insights for more interpretable, more trustworthy and informed decision-making. The dataset included high-resolution RGB and color-rendered images from 79 patients, along with clinical and demographic data such as age, gender, oral hygiene index, and gingival bleeding. Features were extracted using Pyradiomics from manually segmented plaque regions. Feature selection was performed using Recursive Feature Elimination (RFE), followed by the application of multiple machine learning algorithms for classification. Results revealed that the blue color channel of plaque areas achieved the highest accuracy (80%) in distinguishing moderate from severe periodontitis, emphasizing the different fluorescent properties of plaque and teeth, induced by blue light. Notably, sphericity emerged as a key shape feature: as sphericity increased, predictions leaned toward less severe periodontitis, which is consistent with the clinical observation that more circular plaque areas tend to indicate localized and less extensive disease. These clinically relevant insights not only validate the proposed approach applied to intraoral photographic images but also demonstrate its reliability, paving the way for new, more explainable diagnostic methodologies.