U-NET APPLIED TO BONE SEGMENTATION ON COMPUTED MICROTOMOGRAPHIES OBTAINED BY SYNCHROTRON RADIATION FOR HISTOMORPHOMETRIC ANALYSES

• Autor(a) principal: Souza Premoli Pinto de Oliveira, Vitor
• Data de Publicação: 2023
• Outros Autores: Destefani Stefanato, Eduardo, Jorge Gomes Pinheiro, Christiano, Cély Rodrigues Barroso, Regina, Alvarenga de Moura Meneses, Anderson
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Revista Interdisciplinar de Pesquisa em Engenharia
• Texto Completo: https://periodicos.unb.br/index.php/ripe/article/view/46854
• topic: U-Net; Segmentation; Computerized Microtomography; Histomorphometry; Synchrotron Radiation; U-Net; Segmentaçao; Microtomografia Computadorizada; Radiação Síncrotron

Actually, artificial intelligence (AI) participates increasingly in the elaboration of biomedical diagnoses. Clinical applications have used deep learning (DP) methods in the segmentation process, helping in the early treatment of diseases. Based on this principle, this work proposes, via Deep Neural Network (DNN), U-Net, to segment images of rat tibia, the main idea was to use AI architectures added to the image quantification technique, bone histomorphometry. To obtain the images, it was used the non-destructive technique of Computerized Microtomography obtained by X-rays from Synchrotron Radiation (µTC-RS). The initial objective was to enable models to eliminate marrow and other artifacts, leaving only bone; the final objective was to contribute to the state of the art in the use of PA-based methods in contrast to traditional segmentation methods, seeking to apply them to biomedical images. In this study, the developed models resulted in an average of approximately 90% for the Sørensen-Dice coefficient metric, demonstrating a high replicability rate.