Development and Validation of an Artificial Intelligence-based Mobile Application for Periodontal Diagnosis

Periodontal diseases affect more than half of the global adult population, yet their clinical diagnosis remains operatordependent and subject to inter-examiner variability. This retrospective diagnostic accuracy study aimed to develop and internally validate PerioScan AI, an artificial intelligence-based mobile application for periodontal diagnosis across seven diagnostic categories. Clinical case sheets from 745 patients attending a tertiary dental institution in Chennai, India, were pre-processed using custom Python pipelines. After removal of exact-match duplicate records, 435 unique records were retained and partitioned into a training set (n = 348, 80%) and a held-out test set (n = 87, 20%) by stratified random splitting. Seven classes were represented: periodontally healthy, localised and generalised chronic gingivitis, localised and Generalised Chronic Periodontitis (GCP), and localised and Generalised Aggressive Periodontitis (GAP). A total of 964 clinical features were extracted and one-hot encoded, yielding 3,957 predictor variables, and a Random Forest Classifier was trained and validated against the consensus diagnosis of two calibrated periodontists. Outcomes were overall accuracy, class-level sensitivity, specificity, area under the receiver operating characteristic curve and Cohen's kappa. On the held-out test set the classifier achieved an overall accuracy of 82.2% (95% confidence interval 72.9% to 88.8%) and a macro-averaged area under the curve of 0.780, marginally below the prespecified threshold of 0.80; class-level sensitivity ranged from 75.3% to 82.2%. Agreement between applicationgenerated management recommendations and expert consensus was substantial (Cohen’s kappa 0.80, 95% confidence interval 0.74 to 0.82). PerioScan AI shows promise as a first-generation decision support aid, but prospective multicentre validation incorporating 2017 staging and grading parameters is required before clinical deployment.