Oncoscreening and early detection of cancer of the oral mucosa by analyzing photographic images using neural networks

Objective. To create a new clinically and cost-effective technique that improves the effectiveness of cancer screening and early detection of cancer of the oral mucosa and the red border of the lips. Material and methods. A clinical examination of 1.560 patients aged 18 to 85 years with dental diseases was conducted, clinical, special (photometry, autofluorescence stomatoscopy), statistical methods, and machine learning methods were used. Microsoft Excel Office 365 (USA), product version 16.0.4615.1000. Results and discussion. A method for analyzing photographic images using neural networks has been developed, and a “Photobank” has been created: a repository and WEB service for early detection from photographs using a neural network with the ability to replenish it, train and retrain neural networks to improve the accuracy of diagnosing precancerous and oncological diseases of the oral mucosa and red lip rim, and create a telemedicine service. Architectures of 7 databases of photographs of patients with diseases of the oral mucosa and the red border of the lips are registered. The development and approval of the test program and methodology, confirmation of the required characteristics, determination of the nomenclature code of the multifunctional camera type, conducting tests in clinical settings, and collecting photographs for nosology datasets were carried out. Conclusions. 1. The created electronic databases of photos of pathological elements of the mucous membrane allow the neural network to be trained to automatically distinguish intact and pathologically altered areas. 2. The Photo Bank web platform is suitable for storing photos of pathological elements, providing a system for supporting medical decision-making during remote consultations for the detection of precancerous diseases, and for telemedicine consultations between dentists and experts on detected cases of oncological vigilance. 3. The created algorithms for collecting data and the protocol for taking photos of the oral mucosa and red lip border with anchor tags minimize subjective factors and allow for the creation of standardized photo databases for training artificial intelligence and consulting using neural networks.