ABSTRACT
BACKGROUND: At present, clinical use of MRI in Alzheimer's disease (AD) is mostly focused on the assessment of brain atrophy, namely in the hippocampal region. Despite this, multiple biomarkers reflecting structural and functional brain connectivity changes have shown promising results in the assessment of AD. To help identify the most relevant ones that may stand a chance of being used in clinical practice, we compared multiple biomarker in terms of their value to discriminate AD from healthy controls and analyzed their age dependency. METHODS: 20 AD patients and 20 matched controls underwent MRI-scanning (3T GE), including T1-weighted, diffusion-MRI, and resting-state-fMRI (rsfMRI). Whole brain, white matter, gray matter, cortical gray matter and hippocampi volumes were measured using icobrain. rsfMRI between regions of the default-mode-network (DMN) was assessed using group independent-component-analysis. Median diffusivity and kurtosis were determined in gray and white-matter. DTI data was used to evaluate pairwise structural connectivity between lobar regions and the hippocampi.Logistic-Regression and Random-Forest models were trained to classify AD-status based on, respectively different isolated features and age, and feature-groups combined with age. RESULTS: Hippocampal features, features reflecting the functional connectivity between the medial-Pre-Frontal-Cortex (mPFC) and the posterior regions of the DMN, and structural interhemispheric frontal connectivity showed the strongest differences between AD-patients and controls. Structural interhemispheric parietal connectivity, structural connectivity between the parietal lobe and hippocampus in the right hemisphere, and mPFC-DMN-features, showed only an association with AD-status (p < 0.05) but not with age. Hippocampi volumes showed an association both with age and AD-status (p < 0.05).Smallest-hippocampus-volume was the most discriminative feature. The best performance (accuracy:0.74, sensitivity:0.74, specificity:0.74) was obtained with an RF-model combining the best feature from each feature-group (smallest hippocampus volume, WM volume, median GM MD, lTPJ-mPFC connectivity and structural interhemispheric frontal connectivity) and age. CONCLUSIONS: Brain connectivity changes caused by AD are reflected in multiple MRI-biomarkers. Decline in both the functional DMN-connectivity and the parietal interhemispheric structural connectivity may assist sepparating healthy-aging driven changes from AD, complementing hippocampal volumes which are affected by both aging and AD.
ABSTRACT
BACKGROUND: Whole brain atrophy (WBA) estimates in multiple sclerosis (MS) correlate more robustly with clinical disability than traditional, lesion-based metrics. We compare Structural Image Evaluation using Normalisation of Atrophy (SIENA) with the icobrain longitudinal pipeline (icobrain long), for assessment of longitudinal WBA in MS patients. METHODS: Magnetic resonance imaging (MRI) scan pairs [1.05 (±0.15) year separation] from 102 MS patients were acquired on the same 3T scanner. Three-dimensional (3D) T1-weighted and two-dimensional (2D)/3D fluid-attenuated inversion-recovery sequences were analysed. Percentage brain volume change (PBVC) measurements were calculated using SIENA and icobrain long. Statistical correlation, agreement and consistency between methods was evaluated; MRI brain volumetric and clinical data were compared. The proportion of the cohort with annualized brain volume loss (aBVL) rates ⩾ 0.4%, ⩾0.8% and ⩾0.94% were calculated. No evidence of disease activity (NEDA) 3 and NEDA 4 were also determined. RESULTS: Mean annualized PBVC was -0.59 (±0.65)% and -0.64 (±0.73)% as measured by icobrain long and SIENA. icobrain long and SIENA-measured annualized PBVC correlated strongly, r = 0.805 (p < 0.001), and the agreement [intraclass correlation coefficient (ICC) 0.800] and consistency (ICC 0.801) were excellent. Weak correlations were found between MRI metrics and Expanded Disability Status Scale scores. Over half the cohort had aBVL ⩾ 0.4%, approximately a third ⩾0.8%, and aBVL was ⩾0.94% in 28.43% and 23.53% using SIENA and icobrain long, respectively. NEDA 3 was achieved in 35.29%, and NEDA 4 in 15.69% and 16.67% of the cohort, using SIENA and icobrain long to derive PBVC, respectively. DISCUSSION: icobrain long quantified longitudinal WBA with a strong level of statistical agreement and consistency compared to SIENA in this real-world MS population. Utility of WBA measures in individuals remains challenging, but show promise as biomarkers of neurodegeneration in MS clinical practice. Optimization of MRI analysis algorithms/techniques are needed to allow reliable use in individuals. Increased levels of automation will enable more rapid clinical translation.
