Development and clinical validation of a point-of-care cytology screening tool for oral potentially malignant disorders and oral squamous cell carcinoma using a deep learning convolutional neural network: protocol for a diagnostic accuracy study

Introduction

Oral potentially malignant disorders (OPMDs) and oral squamous cell carcinoma (OSCC) present a remarkable public health challenge worldwide. They are the leading causes of cancer-related morbidity and mortality in low-resource regions. In rural and underserved populations, access to diagnostic facilities is limited, and delays in confirmation often result in late-stage presentations. Field screening camps frequently serve as the first and sometimes only point of contact for many at-risk individuals; however, conventional cytology relies on laboratory infrastructure and specialist review, making same-visit diagnoses unfeasible. Recent advances in artificial intelligence-driven systems have created opportunities for point-of-care (PoC) diagnosis, enabling real-time disease detection and triage directly in community settings. This study aims to develop and clinically validate a convolutional neural network (CNN)-enabled PoC cytology device for real-time screening of OPMDs such as leukoplakia, oral submucous fibrosis, erythroplakia and OSCC in community field settings, following Standards for Reporting Diagnostic Accuracy Studies 2015 (STARD) and Standard Protocol Items: Recommendations for Interventional Trials 2013 (SPIRIT) guidelines.

Methods and analysis

This diagnostic accuracy study will evaluate an artificial intelligence (AI)-based PoC cytology screening tool for OPMDs and OSCC. A total of 900 participants (360 retrospective, 540 prospective) will be included to ensure adequate representation of normal, OPMDs and OSCC cases for both model development and clinical validation. Retrospective smears will be used to train and internally validate a deep learning CNN to classify smears as low-risk or high-risk. All slides will be independently reviewed by two blinded cytopathologists, with a third adjudicator resolving disagreements; inter-rater reliability will be assessed using Cohen’s kappa. Prospective validation will be evaluated in patients with clinically suspicious lesions against AI predictions with cytopathologist review and histopathology when available. Diagnostic performance will be assessed using sensitivity, specificity, positive and negative predictive values and area under the receiver operating characteristic curve, which will be calculated. Secondary outcomes are concordance with expert cytology, turn-around time and operational feasibility.

Ethics and dissemination

The protocol adheres to the Central Ethics Committee on Human Research ethical guidelines and has received institutional ethical approval from Datta Meghe Institute of Higher Education and Research (Deemed University) with the reference no. DMIHER (DU)/IEC/2024/18. Written informed consent will be obtained before recruitment.

The results will be published in peer-reviewed scientific journals and presented at national and international conferences.

Trial registration number

CTRI/2025/01/079408.