Estimating the mechanical competence parameter of the trabecular bone: A neural network approach

• Autor(a) principal: Filletti, Érica Regina [UNESP]
• Data de Publicação: 2016
• Outros Autores: Roque, Waldir Leite
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1590/2446-4740.05615; http://hdl.handle.net/11449/173359
• topic: Artificial neural network; Machine learning; Mechanical competence; Osteoporosis; Trabecular bone

Introduction: The mechanical competence parameter (MCP) of the trabecular bone is a parameter that merges the volume fraction, connectivity, tortuosity and Young modulus of elasticity, to provide a single measure of the trabecular bone structural quality. Methods: As the MCP is estimated for 3D images and the Young modulus simulations are quite consuming, in this paper, an alternative approach to estimate the MCP based on artificial neural network (ANN) is discussed considering as the training set a group of 23 in vitro vertebrae and 12 distal radius samples obtained by microcomputed tomography (μCT), and 83 in vivo distal radius magnetic resonance image samples (MRI). Results: It is shown that the ANN was able to predict with very high accuracy the MCP for 29 new samples, being 6 vertebrae and 3 distal radius bones by μCT and 20 distal radius bone by MRI. Conclusion: There is a strong correlation (R2= 0.97) between both techniques and, despite the small number of testing samples, the Bland-Altman analysis shows that ANN is within the limits of agreement to estimate the MCP.