Faculty, Staff and Student Publications

Language

English

Publication Date

6-17-2026

Journal

Journal of Clinical Medicine

DOI

10.3390/jcm15124692

PMID

42355860

PMCID

PMC13301221

PubMedCentral® Posted Date

6-17-2026

PubMedCentral® Full Text Version

Post-print

Abstract

Background/Objectives: Oral potentially malignant disorders (OPMDs) require accurate risk stratification to identify patients at the highest risk for severe oral epithelial dysplasia (OED) or oral squamous cell carcinoma (OSCC). We developed and internally validated the oral cancer numerical index (OCNI), a quantitative risk score derived from clinical features and deep learning-based brush cytology measurements.

Methods: This retrospective model development and internal validation study was conducted using data from the multicenter Grand Opportunity study. Prospectively recruited subjects with OPMD with complete data were divided at the subject level into a training set (n = 384) and a holdout test set (n = 164) using a 70:30 diagnosis-stratified split. The primary endpoint was severe OED or OSCC versus benign diagnoses, and mild and moderate OED. Predictors included age, sex, tobacco history, lesion color, lesion size, multiple lesions, ulcerative morphology, and the percentages of differentiated squamous epithelial and small round cells derived from deep learning-based cytology. Prespecified rule-out and rule-in thresholds were selected in the training set to target 90% sensitivity and 90% specificity, respectively, and then applied to the holdout test set.

Results: At the prespecified rule-out threshold (OCNI ≤ 37.6), sensitivity was 92% and negative predictive value was 97%. At the rule-in threshold (OCNI > 60.0), specificity was 89% and positive predictive value was 67%. Calibration was good in the holdout set (intercept, -0.07; slope, 1.13; Hosmer-Lemeshow p = 0.36), and OCNI increased significantly with worsening histopathologic severity.

Conclusions: OCNI provided an objective, clinically interpretable estimate of risk for severe OED or OSCC, with strong rule-out and rule-in performance and good calibration. These findings support further external validation of OCNI as an adjunctive tool for oral lesion risk stratification.

Keywords

oral potentially malignant disorders, oral epithelial dysplasia, oral squamous cell carcinoma, cytology, deep learning, artificial intelligence, intelligent cytology microfluidics

Published Open-Access

yes

Download

Included in

Dentistry Commons

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.