Who will develop Alzheimer’s disease? New insights from multimodal neuroimaging
Letizia Casiraghi1,2, Fulvia Palesi2,3, Gloria Castellazzi2,4, Andrea De Rinaldis2,4, Elena Sinforiani5, Claudia Angela Michela Gandini Wheeler-­Kingshott 2,6, Egidio D'Angelo1,2, and Carol Di Perri2

1Department of Brain and Behavioral Sciences, University of Pavia, Pavia, Italy, 2Brain Connectivity Center, C. Mondino National Neurological Institute, Pavia, Italy, 3Department of Physics, University of Pavia, Pavia, Italy, 4Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy, 5Neurology Unit, C. Mondino National Neurological Institute, Pavia, Italy, 6NMR Research Unit, Queen Square MS Centre, Department of Neuroinflammation, UCL Institute of Neurology, University College London, London, United Kingdom

Synopsis

Mild cognitive impairment (MCI) is considered a transitional state between healthy controls (HC) and Alzheimer’s disease (AD). This study compares the predictive value of neuropsychological evaluation, structural magnetic resonance imaging, diffusion tensor imaging and resting state functional MRI indices able to identify MCI conversion. AD versus HC and converted MCI (cMCI) versus non-converted MCI (ncMCI) presented different features of differentiation. This result suggests adopting advanced MRI techniques to investigate early alterations. Due to the clinical heterogeneity of MCI patients, considering cMCI as AD-like and ncMCI as HC might be inappropriate when attempting to distinguishing between cMCI and non-converted MCI.

Purpose

Mild Cognitive Impairment (MCI) clinically is considered as a transitional state between healthy controls (HC) and Alzheimer Disease (AD)1. However, the annual conversion rate of MCI to AD is approximately 12%2. Therefore, the recognition of MCI who will develop (converters MCI, cMCI) or not (non-converters MCI, ncMCI) to AD and identification of optimal markers of disease progression is of undeniable importance. To the best of our knowledge no one has compared, on a single dataset, the predictive value of neuropsychological evaluation (NPS), structural magnetic resonance imaging (sMRI), diffusion tensor imaging (DTI) and resting state functional MRI (rsfMRI). Therefore the aim of this study is to define the predictive value of each metric in identifying MCI that converted to AD within 2 years.

Methods

98 subjects: 36 AD (19 f, 71.8±5.8 years, 7.0±2.6 years of education, MMSE 21.7±3.1), 31 MCI (19 f, 72.5±5.7 years, 7.9±4.2 years of education, MMSE 24.8±2.7), and 31 HC (21 f, 68.2±6.8 years, 7.9±3.9 years of education, MMSE 28.5±1.3) underwent NPS and MRI on a Philips Intera 1.5T scanner with an 8-channel head coil. Volumetric T1-weighted images (3DT1 FFE, TR/TE = 8.6/4 ms, 170 sagittal slices, slice thickness = 1.2 mm, in-plane resolution = 1.25×1.25 mm2), DTI (SE-EPI, TR/TE = 11800/70 ms, 2.5 mm isotropic voxel, 15 non-collinear directions, b-values = 900 s/mm2) and rsfMRI (FE­EPI, TR/TE = 3000/60 ms, voxel size = 2.2x2.2x4 mm3, 100 volumes) sequences were acquired. At follow-up (12±8 months) 12 MCI converted to AD. Grey and white matter brain areas showing significant differences among groups were determined using Voxel Based Morphometry (sMRI) – VBM3, Tract-based spatial statistics (DTI) – TBSS4 and dual regression (rsfMRI)5 analysis. Metrics for grey matter volumes, diffusivity and anisotropy, and functional connectivity (FC) values from the highlighted areas were used as input features for automatic classification implemented in Orange6. Using a leave one out cross validation approach and testing different classification algorithms (Logistic Regression, K Nearest Neighbours, Support Vector Machine, and Classification Tree) we assessed the predictive power of each index in differentiating the pathological groups by testing the classifier accuracy (ACC) on NPS, sMRI, DTI and rsfMRI separately. The relationship among the obtained features and NPS was then verified using linear regression analysis with SPSS 21.07.

Results

Our findings confirmed previous literature on AD, highlighting that the best classification performance results from NPS and sMRI data [Fig.1]. With regard to the cMCI group, the best classification performance was obtained when considering rsfMRI and DTI metrics [Fig.2]. In classifying AD versus HC the combined use of 6 features from NPS and sMRI [Tab.1] reached ACC of 97%, sensitivity (SEN) of 94%, specificity (SPE) of 100% and area under curve (AUC) of 98%. Furthermore, linear regression analysis revealed statistically significant relationships between right hippocampal volume and several NPS: memory prose (p<0.001, R2=0.42), Rey-Osterrieth complex figure delayed recall (p<0.001, R2=0.39), TMT-A (p<0.001, R2=0.19), TMT-B (p<0.001, R2=0.32), and phonemic verbal fluency (p<0.006, R2=0.12). On the other hand, when classifying cMCI versus ncMCI 9 features [Tab.2] from all investigated techniques, were needed to reach ACC of 77% (SE = 66%, SPE = 84% and AUC = 74%). Finally, in MCI linear regression analysis showed statistically significant relationship between Corsi block task and two volumetric measures: right superior parietal lobe volume (p=0.025, R2=0.18), and right medial temporal lobe volume (p=0.031, R2=0.17).

Discussion and conclusion

Our study is in line with previous work emphasizing the importance of NPS and conventional sMRI data in confirming AD. The final AD classification model resulted highly accurate, sensitive and specific indeed. The observed relationship between atrophy and cognitive deficits in both AD and cMCI stresses the specificity of verbal memory and attention system deficits in AD, as well as an impaired visuospatial memory system in cMCI. Since AD versus HC and cMCI versus ncMCI presented different features of differentiation, our results suggests the need for adopting advanced techniques (fMRI and DTI) to properly investigate early cerebral alterations that wouldn’t be detected using conventional approaches only. We argue that, considering cMCI as AD-like patients and ncMCI as HC-like subjects might be inappropriate when attempting to distinguish between cMCI and ncMCI. Further investigations are needed to assess the heterogeneity of this diagnostic entity; bigger and more homogeneous cohort of patients could guarantee the improvement of the classifier’s sensitivity in detecting the signs of MCI conversion.

Acknowledgements

Brain Connectivity Center (BCC) - C. Mondino National Neurological Institute, University of Pavia and Italian Ministry of Health (GR-2009-1575236) for fundings.

References

1. Holtzman, D.M. (2011), Alzheimer’s Disease: the challenge of the second century, Sci Transl Med., Vol. 3, n. 77, 77srl.

2. Mansbach, W.E. (2015), Mild cognitive impairment (MCI) in long-term care patients: subtype classification and occurrence. Aging Ment Health, p. 1-6.

3. Ashburner, J. and Friston, K.J. (2001), Why voxel-based morphometry should be used. Neuroimage. 14(6): p. 1238-43.

4. Smith, S.M. (2006), Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data. NeuroImage, 31:1487-1505.

5. Beckmann, C.F. (2009), Group comparison of resting-state FMRI data using multi-subject ICA and dual regression. OHBM.

6. Demsar, J. (2013), Orange: Data Mining Toolbox in Python. Journal of Machine Learning Research 14(Aug) 2349−2353.

7. IBM Corp. Released 2012. IBM SPSS Statistics for Mac, Version 21.0. Armonk, NY: IBM Corp.

Figures

Fig. 1: AD classification based on the logistic regression classifier. The classification accuracy is reported along y and the selected data origin is reported along x. These results confirm previous knowledge on AD markers.

Figure 2: cMCI classification based on the logistic regression classifier. The classification accuracy is reported along y and the selected data type is reported along x. This result highlights the usefulness of multimodal neuroimaging and advanced techniques (DTI and rsfMRI) for mild pathological states classification.

Table 1: The best features (attributes) for AD classification are 5 neuropsychological scores and the right hippocampal volume. Verbal memory and attention system deficits seem to be AD specific. Writing colours correspond to bar colours of Fig.1.

Table 2: cMCI classification needs rsfMRI, DTI, NPS and sMRI data for a successful classification. The verbal/visuospatial memory system deficit, assessed by NPS and VBM, is specific, but not sufficient. Additional information on functional connectivity, fractional anisotropy (FA) and mean diffusivity (MD) are reported as conversion indices. Writing colours correspond to bar colours of Fig.2.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0630