Fast and smart segmentation of paraspinal muscles in magnetic resonance imaging with CleverSeg

Autor(es)

Ramos, Jonathan S. ; Cazzolato, Mirela T. ; Faical, Bruno S. ; Linares, Oscar A. C. ; Nogueira-Barbosa, Marcello H. ; Traina, Caetano, Jr. ; Traina, Agma J. M. ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: 2019 32ND SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI); v. N/A, p. 8-pg., 2019-01-01.

Resumo

Magnetic Resonance Imaging (MRI) is a noninvasive technique, which has been employed to detect and diagnose many spine pathologies. In a Computer-Aided Diagnosis (CAD) context, the segmentation of the paraspinal musculature from MRI may support measurement, quantification, and analysis of muscle-related pathologies. Current semi-automatic segmentation techniques require too much time from the physicians to annotate all slices in the exams. In this work, we focus on minimizing the time spent on manual annotation as well as on the overall segmentation processing time. We use the mean absolute error between slices aiming at minimizing the number of annotated slices in each exam. Moreover, we optimize the manual annotation time by estimating the inside annotation based on the outside annotation, while the competitors demand the annotation of inside and outside annotation (seeds). The experimental evaluation shows that our proposed approach is able to speed up the manual annotation process in up to 50% by annotating only a few representative slices, without loss of accuracy. By annotating only the outside region, the process can be further speed up by another 50%, reducing the total time to only 25% of the previously required. Thus, the total time spent on manual annotation is reduced by up to 75%, and, since human interaction is greatly diminished, allows a more productive and less tiresome activity. Despite that, our proposed CleverSeg method presented accuracy similar to or better than the competitors, while managing a faster processing time. (AU)

• Processo FAPESP: 18/24414-2 - Ambiente para integração de técnicas para a extração de características e bases de dados complexos para o projeto MIVisBD
• Beneficiário: Mirela Teixeira Cazzolato
• Modalidade de apoio: Bolsas no Brasil - Programa Capacitação - Treinamento Técnico
• Processo FAPESP: 16/17078-0 - Mineração, indexação e visualização de Big Data no contexto de sistemas de apoio à decisão clínica (MIVisBD)
• Beneficiário: Agma Juci Machado Traina
• Modalidade de apoio: Auxílio à Pesquisa - Temático
• Processo FAPESP: 17/23780-2 - Recuperação por conteúdo de imagens médicas para apoio a decisão clínica usando a abordagem radiômica
• Beneficiário: Jonathan da Silva Ramos
• Modalidade de apoio: Bolsas no Brasil - Doutorado
• Processo FAPESP: 18/06228-7 - Detecção de padrões e anomalias em dados médicos usando Modelagem Matemática
• Beneficiário: Bruno Squizato Faiçal
• Modalidade de apoio: Bolsas no Brasil - Pós-Doutorado