Study of longitudinal neurodegeneration biomarkers to support the early diagnosis of alzheimer’s disease
- Autores: Giovana Elizabeth Gavidia Bovadilla
- Directores de la Tesis: Alexandre Perera Lluna (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2018
- Idioma: español
- Tribunal Calificador de la Tesis: Pere Caminal Magrans (presid.), Marc Via García (secret.), Sergi Bermúdez i Badía (voc.)
- Programa de doctorado: Programa de Doctorado en Ingeniería Biomédica por la Universidad Politécnica de Catalunya y la Universidad de Zaragoza
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
Alzheimer’s Disease (AD) is a progressive and neurodegenerative disorder characterized by pathological brain changes starting several years before clinical symptoms appear. Earlier and accurate identification of those brain structures changes can help to improve diagnosis and monitoring, allowing that future treatments target the disease in its earliest stages, before irreversible brain damage or mental decline takes place. The brain of AD subjects shrinks significantly as the disease progress. Furthermore, ageing is the major risk factor for sporadic AD, older brains being more susceptible than young or middle-aged ones. However, seemingly healthy elderly brains lose matter in regions related to AD. Likewise, similar changes can also be found in subjects having mild cognitive impairment (MCI), which is a symptomatic pre-dementia phase of AD. This work proposes two methods based on statistical learning methods, which are focused on characterising the ageing-related changes in brain structures of healthy elderly controls (HC), MCI and AD subjects, and addressing the estimation of the current diagnosis (ECD) of HC, MCI and AD, as well as the prediction of future diagnosis (PFD) of these groups mainly focused on the early diagnosis of conversion from MCI to AD. Data correspond to longitudinal neurodegeneration measurements from Magnetic Resonance Imaging (MRI) images. These biomarkers were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Open Access Series of Imaging Studies (OASIS). ADNI data includes MRI biomarkers available at a 5-year follow up on HC, MCI and AD subjects, while OASIS data only includes biomarkers measured at baseline on HC and AD. In the first method, called M-res, variant (vr) and quasi-variant (qvr) biomarkers were identified on HC subjects by using a Linear Mixed Effects (LME) approach on males and females, separately. Then, we built an ageing-based null model, which would characterise the normal atrophy and growth patterns of vr and qvr biomarkers, as well as the correlation between them. By using the null model on those subjects who had been clinically diagnosed as HC, MCI or AD, normal age-related changes were estimated, and then, their deviation scores (residuals) from the observed MRI-based biomarkers were computed. In contrast to M-res, the second method, called M-raw, is focused on directly analyzing the raw MRI-based biomarkers values stratified by five-year age groups. M-raw includes a differential diagnosis-specific feature selection (FS) method, which is applied before classification. In both methods, the differential diagnosis problem was addressed by building Support Vector Machines (SVM) models to carry out three main experiments—AD vs. HC, MCI vs. HC, and AD vs. MCI. In M-res, the SVM models were trained by using as input the residuals computed for the vr biomarkers plus the age, whereas in M-raw, we used the pool of selected features plus age, gender and years of education. The advancement of early disease prediction was calculated as the average number of years advanced in the PFD of the subjects concerning the last known clinical diagnosis. Finally, the ability of both methods to correctly discriminate AD vs. HC subjects was evaluated and compared by testing them on OASIS subjects observed at baseline. Results confirm accelerated or reduced estimates of decline in all cortical biomarkers with increasing age and a frontotemporal pattern of atrophy in HC subjects, as well as in MCI and AD. Regarding the ECD problem, all SVM models obtained better results than comparable methods in the literature for most classification quality indicators, especially on AD vs. HC. Both methods also improve the PFD given the current clinical tests, both in prediction quality indicators and the amount of time by which the diagnosis is advanced.
