Computerized tools : a substitute or a supplement when diagnosing Alzheimer's disease?

Alzheimer’s disease (AD) is the most common form of dementia in the elderly characterized by difficulties in memory, disturbances in language, changes in behavior, and impairments in daily life activities. By the time cognitive impairment manifests, substantial synaptic and neuronal degeneration has already occurred. Therefore, patients need to be diagnosed as early as possible at a preclinical or presymptomatic stage. This will be important when disease-modifying treatments exist in the future.

The main focus of this thesis is on the study of structural neuroimaging in AD and in prodromal stages of the disease. We emphasize the use of statistical learning for the analysis of structural neuroimaging data to achieve individual prediction of disease status and conversion from prodromal stages. The main aims of the thesis were to develop and validate computerized tools to identify patterns of atrophy with the potential of becoming markers of AD pathology using structural magnetic resonance imaging (sMRI) data and to develop a segmentation tool for Computed Tomography (CT).

Using automated neuroanatomical software we measured multiple brain structures that were given to statistical learning techniques to create discriminative models for prediction of presence of disease and conversion from prodromal stages. Building statistical models based on sMRI data we investigated optimal normalization strategies for the combination of structural measures such as cortical thickness, cortical and subcortical volumes (Study I). A baseline model was created based on the optimal normalization strategy and combination of structural measures. This model was used to compare the discrimination ability of different statistical learning algorithms (decision trees, artificial neural networks, support vector machines and orthogonal partial least squares (OPLS)). Additionally, the addition of age, years of education and APOE phenotype was added to the baseline model to assess the impact on discrimination ability (Study II). The OPLS classification algorithm was trained on the baseline model to produce a structural index reflecting information about AD-like patterns of atrophy from each individual’s sMRI data. Additional longitudinal information at one-year follow-up was used to characterize the temporal evolution of the derived index (Study III). Since total intracranial volume (ICV) remains a morphological measure of interest and CT is today widely used in routine clinical investigations, we developed and validated an automated segmentation algorithm to estimate ICV from CT scans (Study IV).

We believe computerized tools (automated neuroimaging software and statistical discriminative algorithms) have significantly enriched our knowledge and understanding of associated neurodegenerative pathology, its effects on cognition and interaction with age. These tools were mainly developed for research purposes but we believe all accumulated knowledge and insights could be translated into clinical settings, however, that is a challenge that remains open for future studies.

List of scientific papers

I. Westman, E., Aguilar, C., et al., 2013. Regional magnetic resonance imaging measures for multivariate analysis in Alzheimer's disease and mild cognitive impairment. Brain Topography, Volume 26, Issue 1, p. 9-23.
https://doi.org/10.1007/s10548-012-0246-x

II. Aguilar, C., Westman, E., et al., 2013. Different multivariate techniques for automated classification of MRI data in Alzheimer’s disease and mild cognitive impairment. Psychiatry Research: Neuroimaging, Volume 212, Issue 2, p. 89-98.
https://doi.org/10.1016/j.pscychresns.2012.11.005

III. Aguilar, C., Muehlboeck, S., et al., 2014. Application of a MRI based index to longitudinal atrophy change in Alzheimer disease, mild cognitive impairment and healthy older individuals in the AddNeuroMed cohort. Frontiers in Aging Neuroscience.
https://doi.org/10.3389/fnagi.2014.00145

IV. Aguilar, C., Edholm, K., et al. 2015. Automated CT-based segmentation and quantification of total intracranial volume. Eur Radiol. 2015 Apr 16.
https://doi.org/10.1007/s00330-015-3747-7