Hybrid LSM-based Image Segmentation and Analysis of Morphological Variations of the Brainstem in Alzheimer MR Images

Abstract

The current diagnostic evaluation criteria for Alzheimer’s disease (AD) are restricted to cognitive deficits, and less consideration is given to AD-related noncognitive, behavioral, and psychological symptoms of dementia. Vascular deterioration in AD subjects manifests neuropsychiatric symptoms (NPS). Early anxiety and depression are associated with cognitive decline. The brainstem undergoes deformation during the initiation of AD due to NPS. In this work, structural deformation of the brainstem is analyzed using Lattice Boltzmann criterion-based hybrid level set method (LSM) and fourth-order diffusion filtering. Contour and skeletal-based shape measures are extracted and analyzed using multilayer perceptron (MLP). Results show that the fourth-order diffusion filter provides strong edges, which aids in accurate segmentation. The Lattice Boltzmann criterion-based hybrid LSM is able to delineate the whole brainstem. Hough transform statistics (HTS) obtained from the segmented brainstem describe the geometric properties of the boundary of the brainstem. Eigen space values identify the skeletal variations of the brainstem for normal, Mild Cognitive Impairment (MCI), and AD subjects. Both HTS and eigen space features are found to represent better differentiation between normal, MCI, and AD subjects with a significance of p＜0.05. It is observed that Hough transform–based statistical values have the ability to better discriminate the structural deformity of the brainstem with classification accuracies of 73%, 68%, and 71% for normal–MCI, MCI–AD, and normal–AD classifications, respectively, using MLP. Thus, this analysis identifies the presence of structural abnormality in the brainstem during the progression of AD, and therefore, this analysis could aid in early diagnosis of AD.