Leandro Beltrachini, Matteo De Marco, Zeike A. Taylor, Jyrki Lotjonen, Alejandro F. Frangi and Annalena Venneri Pages 592 - 603 ( 12 )
Background: Resting-state functional magnetic resonance imaging (RS-fMRI) appears as a promising imaging technique to identify early biomarkers of Alzheimer type neurodegeneration, which can be more sensitive to detect the earliest stages of this disease than structural alterations. Recent findings have highlighted interesting patterns of alteration in resting-state activity at the mild cognitive impairment (MCI) prodromal stage of Alzheimer’s disease. However, it has not been established whether RS-fMRI alterations may be of any diagnostic use at the individual patient level and whether parameters derived from RS-fMRI images add any quantitative predictive/classificatory value to standard cognitive tests (CTs). Methods: We computed a set of 444 features based on RS-fMRI and used 21 variables obtained from a neuropsychological assessment battery of tests in 29 MCI patients and 21 healthy controls. We used these indices to evaluate their impact on MCI/healthy control classification using machine learning algorithms and a 10-fold cross validation analysis. Results: A classification accuracy (sensitivity/ specificity/area under curve/positive predictive value/negative predictive value) of 0.9559 (0.9620/0.9470/ 0.9517/0.9720/0.9628) was achieved when using both sets of indices. There was a statistically significant improvement over the use of CTs only, highlighting the superior classificatory role of RS-fMRI. Conclusions: RS-fMRI provides complementary information to CTs for MCI-patient/healthy control individual classification.
Accuracy, AD, biomarker, classification, mild cognitive impairment, MCI, neurodegeneration.
Centre for Computational Imaging & Simulation Technologies in Biomedicine (CISTIB), Department of Electronic and Electrical Engineering, The University of Sheffield, Pam Liversidge Building, Mappin Street, Sheffield, S1 3JD, UK.