Detecting Alzheimer's Disease based on Structural Region Analysis using a 3D Shape Descriptor

  • Kauê T. N. Duarte Unicamp
  • David Gobbi UCalgary
  • Richard Frayne UCalgary
  • Marco Antônio Garcia Carvalho Unicamp


Alzheimer's disease (AD) is a common neurodegenerative dementia that affects older people. Changes in behavior and cognition are the most common characteristics of this disease and are associated with changes in brain structure. Techniques focusing on brain shape have been recently proposed to quantify and understand these changes. One challenge when examining AD is that each anatomical region may have a unique role in and time course for brain deterioration, requiring a whole-brain method that is capable of individual (or regional) analyses at different disease stages. We propose to apply the scale-invariant heat kernel signature descriptor to magnetic resonance brain images in order to evaluate regional shape features across different brain regions. We measured the shape feature similarity in 500 subjects, equally divided across five progressive, disease-based stages. The shape analysis provided a complementary perspective to whole-brain analysis, due to the capability of identifying how different structures degenerate at different rates in the brain. In total, a group of 99 distinct brain regions belonging to cortical and deep gray matter were analyzed across the five disease stages. Preliminary assessment of shape-based analysis of key brain regions demonstrated that SIHKS was predictive of disease stage and disease progression.

Palavras-chave: shape descriptor, alzheimer, brain analysis, image segmentation, image processing, disease progression, magnetic resonance imaging
Como Citar

Selecione um Formato
DUARTE, Kauê T. N.; GOBBI, David; FRAYNE, Richard; CARVALHO, Marco Antônio Garcia. Detecting Alzheimer's Disease based on Structural Region Analysis using a 3D Shape Descriptor. In: CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), 33. , 2020, Evento Online. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2020 . p. 212-219.