Meeting Banner
Abstract #3764

T2-Weighted MRI Increases Machine Learning Accuracy in Alzheimer's Disease

L. Z. Diaz-de-Grenu1, G. B. Williams1, J. Acosta-Cabronero1, J. M. Pereira1, P. J. Nestor1

1Clinical Neurosciences, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom

Support vector machines (SVM) using T1-weighted images offer moderate accuracy in diagnosing Alzheimers disease (AD). Recent work suggests, however, that T2-weighted scans may contain more pathologically relevant information. This study, therefore, tested if T2 data could improve SVM classification. T1 and T2 were compared in whole-brain images and regions of interest (ROI) known to be affected in AD. An ROI focused on the mesial temporal lobe (known to be atrophic) yielded similar accuracy for T1 and T2 (both 88.5%), however adding ROIs known to be rich in beta-amyloid improved diagnostic accuracy (92.3%) but only when using T2 data.

Abstract PDF E-Poster Video

Keywords

according accuracy accurate acquisition addition additional affected aging analyses analyze appear applied array assess assessments atrophy benefits best beyond bias brain channel classification clinical cohort coil collapsed combination common computer concordant confirmed consisted consistent contain contrast contrasts control controls corrected delineated deposition deposits detection diagnosed diagnosis diagnostic direct disease diseases employed enable equipped explored finally format frontal greater handle head health healthy highest highlighted highly hypothesis illustration improve improved inclusion intense intensity isotropic kernel kingdom known latter learning leave lecture light linear lobe machine machines manually mask matched matrix mayo mild modified modulated neurobiology notable note notes occipital overall patients performance plus posterior predominantly previous primarily prior proceedings processed proven pulse recent relatively relevant resolution scanned science scores several skull slices software space spared specifically stripped strong structural studies suggests supervised support system table template temporal together train training transform trio true useful vector volumes volumetric warping whereas whole wrong yielded