Title
Artificial intelligence-assisted oral cytology and salivary biomarker analysis for early detection of oral potentially malignant disorders: An in vitro study
Authors
Akhil Trivedi1, Gopal Srivastava2, Sheetal Grover3, Dewin Arnold4,
Vidyarjan A. Sukhadeve5,* &
Prasad Vagadale6
Affiliation
1Department of Dentistry, Government Medical College, Jalaun (Orai), Uttar Pradesh, India; 2Department of Oral Medicine and Radiology, Eklavya Dental College & Hospital, Kotputli-Behror, Rajasthan, India; 3Department of Conservative Dentistry and Endodontics, MM College of Dental Sciences & Research, Mullana-Ambala, Haryana, India; 4Department ofOral & Maxillofacial Pathology and Microbiology, Geetanjali Dental & Research Institute, Geetanjali University, Udaipur, Rajasthan, India; 5Department of Oral Medicine & Radiology, Yogita Dental College & Hospital, Khed, Ratnagiri, Maharashtra, India;6Department of Oral & Maxillofacial Surgery, S.B. Patil Institute for Dental Sciences & Research, Naubad, Bidar, Karnataka, India.*Corresponding author
Akhil Trivedi - E-mail: akt6600@gmail.com
Gopal Srivastava - E-mail: gopalsrivastava45@gmail.com
Sheetal Grover - E-mail: drgroversheetal@gmail.com
Dewin Arnold - E-mail: dewinarnold08@gmail.com
Vidyarjan A. Sukhadeve - E-mail: vidyarjan99@gmail.com
Prasad Vagadale - E-mail: docprasad2213@gmail.com
Article Type
Research Article
Date
Received September 1, 2026; Revised September 30, 2026; Accepted September 30, 2026, Published September 30, 2026
Abstract
Oral potentially malignant disorders require timely recognition because conventional visual examination and cytology may miss subtle cellular atypia, while single salivary biomarkers frequently show insufficient discriminatory performance. This in vitro diagnostic study evaluated 180 paired oral brush cytology slides and saliva specimens comprising healthy controls, non-dysplastic oral potentially malignant disorders and histopathologically confirmed dysplastic oral potentially malignant disorders. Papanicolaou-stained cytology images were analysed using a convolutional neural network for nuclear segmentation and classification, salivary interleukin-6, interleukin-8, matrix metalloproteinase-9 and lactate dehydrogenase were quantified and multimodal predictions were generated using an extreme-gradient-boosting fusion model. The combined model produced a provisional area under the receiver operating characteristic curve of 0.96, sensitivity of 91.7%, specificity of 93.3% and accuracy of 92.8%, outperforming AI cytology alone and the biomarker panel alone. Thus, integration of objective cytomorphometric information with a compact salivary biomarker panel may provide a useful laboratory adjunct for prioritising oral potentially malignant disorders for biopsy and specialist review.
Keywords
Artificial intelligence (AI), oral cytology, salivary biomarkers, oral potentially malignant disorders (OPMDs), dysplasia, early detection
Citation
Trivedi et al. Bioinformation 22(9): 5569-5573 (2026)
Edited by
Ashwini Dhopte
ISSN
0973-2063
Publisher
License
This is an Open Access article which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. This is distributed under the terms of the Creative Commons Attribution License.
