Banca de QUALIFICAÇÃO: Raiza Querrer Peixoto

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : Raiza Querrer Peixoto
DATE: 05/07/2024
TIME: 08:00
LOCAL: Plataforma Virtual - Microsoft Teams
TITLE:

Deep Learning for osteoporosis screening in panoramic radiographs


KEY WORDS:

Osteoporosis Screening, Panoramic Radiography, Artificial Intelligence, CNN, GradCAM.


PAGES: 100
BIG AREA: Outra
AREA: Defesa
SUMMARY:

This study introduces an advanced method for osteoporosis screening using panoramic radiographs (PRs) through convolutional neural network (CNN) models based on the EfficientNet architecture. It explores two methods: analyzing entire PR images and analyzing cropped images focusing on the mandibular cortical region. A carefully curated image database of 19,295 PRs was evaluated by a calibrated examiner, selecting 750 for the study—579 classified as C1 and 171 as C3. This database creation addresses a significant gap in available resources. The process involved detailed image processing and selection utilizing computer vision tools and data preprocessing techniques. Models EfficientNet B5, B6, and B7 were optimized using transfer learning for classification. Statistical analysis showed that the croppedimage method achieved approximately 98\% accuracy, slightly higher than the fullimage method's 95\% accuracy. The use of Grad-CAM allowed for a detailed visual analysis, highlighting the importance of the mandibular cortical edge in predictions, even when using the full-image method. This finding is significant, suggesting potential for broader applications of full PR images. The study notably contributes to computeraided dental diagnostics by making the training and testing codes public, thus providing an efficient, non-invasive tool for routine osteoporosis screening in PRs.


COMMITTEE MEMBERS:
Externo à Instituição - Hugo Gaêta Araújo - USP
Externo ao Programa - 1613634 - MARCELO ANTONIO MAROTTA - nullPresidente - 2315081 - NILCE SANTOS DE MELO
Interno - 3475960 - PAULO TADEU DE SOUZA FIGUEIREDO
Notícia cadastrada em: 06/06/2024 11:57
SIGAA | Secretaria de Tecnologia da Informação - STI - (61) 3107-0102 | Copyright © 2006-2024 - UFRN - app38_Prod.sigaa32