Sistema de apoyo diagnóstico periodontal con deep learning para las clínicas odontológicas de la Universidad Santo Tomás, 2020: Fase I - Insumos y criterios radiográficos

Abstract

The diagnosis of periodontitis generates a variety of criteria that can lead to the clinician's decision being subjective. Deep learning as machine learning is a computerized tool that allows the information to be handled truthfully, quickly and in a timely manner, in addition to having a high degree of reliability and precision, providing new perspectives for diagnosis, prognosis and treatment planning. To develop a system for digitized periapical radiographic interpretation to support periodontal diagnosis based on Deep Learning: Phase I Radiographic criteria and supplies. The study population made up of a total of 727 digitized diagnostic images (periapical radiographs) stored in the USTA radiological center in the years 2019-2020. Exclusion criteria: Elongated periapical radiographic images, alveolar spaces that house implants. 727 images extracted corresponded to 72 subjects, 45 women (62%) and 27 men (38%). The average number of teeth contributed per person was 24.5 ± 4.4 teeth, on the other hand, the mean of dental loss was 7.3 ± 3.3 teeth. The metrics obtained are similar to other studies, thus we found that the inputs generated in Phase I are correct for use in Phase II, that is, to give continuity, for which only the observations generated in the population balance (in terms of sex distribution) and in the sample size (in terms of radiographic images). This neural network system is developed to identify teeth in their initial phase and will be of great help to the clinician, being able to process many images with specific criteria, supporting the diagnosis efficiently.