Estimativa da idade cerebral e classificação diagnóstica na doença de Alzheimer via campos de deformação em imagens de ressonância magnética

Abstract

Alzheimer’s disease is a progressive neurodegenerative condition and the leading cause of dementia in the elderly population, highlighting the need for computational methods sensitive to subtle brain alterations. This thesis proposes a framework for structural magnetic resonance imaging analysis based on morphological attributes derived from deformation fields estimated by deformable registration of individual images to population templates built using groupwise registration. The approach combines a multi-region analysis of brain structures, 3D Shape Context descriptors, Jacobian determinants, the strain tensor, and machine learning. The first contribution estimates brain age in cognitively normal individuals, achieving a mean absolute error of 1.66 years for males and 1.81 years for females, with coefficients of determination of 0.84 and 0.80, respectively. The second contribution is a multi-region classification framework for distinguishing cognitively normal individuals, subjects with mild cognitive impairment, and patients with Alzheimer’s disease, achieving an AUC of 0.93 and an accuracy of 88.68% for males, and an AUC of 0.94 with an accuracy of 87.89% for females in the classification between cognitively normal individuals and Alzheimer’s patients. The third contribution proposes a graph-based approach for identifying progressive mild cognitive impairment, integrating displacement attributes, Jacobian volumetric variation, and local strain into a fixed population-level topology processed by a Graph Attention Network. This approach achieved an AUC of 0.92, an accuracy of 81.21%, and an F1-score of 0.79 for males, and an AUC of 0.88, an accuracy of 78.41%, and an F1-score of 0.73 for females. Taken together, the results indicate that morphological attributes derived from deformation fields can characterize brain aging, discriminate diagnostic groups, and contribute to differentiating individuals with stable and progressive mild cognitive impairment.