University Medical Center of the Johannes Gutenberg University Mainz
Department of Neurology

Author Of 5 Presentations

Machine Learning/Network Science Poster Presentation

P0004 - Convolutional neural network framework for predicting progression in early MS (ID 1679)

Speakers
Presentation Number
P0004
Presentation Topic
Machine Learning/Network Science

Abstract

Background

Brain tissue damage is closely linked to disability in multiple sclerosis (MS). The localization of white matter (WM) lesions influences the course of the disease.

Objectives

However, the interrelation between lesions topography and cortical atrophy distribution for predicting the clinical disability remains unclear. Use a deep learning neural network framework with the purpose to identify critical co-varying patterns for individualized disease prediction.

Methods

Clinical disability was measured using the Expanded Disability Status Score at baseline and at a one-year follow-up in a cohort of 119 patients with early relapsing-remitting MS and in a replication cohort of 81 patients. Co-varying patterns of cortical atrophy and baseline lesion distribution were extracted by parallel ICA and used as features for constructing a deep learning convolutional neural network. The prediction was conducted for each identified lesion pattern separately using 50% as training cohort and 50% as testing cohort.

Results

In the study cohort, we identified three distinct distribution types of WM lesions (“cerebellar”, “bihemispheric” and “left-lateralized”). The “cerebellar” and “left-lateralized” patterns were reproducibly detected in the second cohort. Each of the patterns predicted to different extents, short-term disability progression, while the “cerebellar” pattern predicting individual disability progression with an 10-fold cross-validation accuracy of above 90% for the Study cohort (95% CI: 88%-94%) and above 85% for the replication cohort (95% CI: 81%-88%) respectively.

Conclusions

These findings highlight that role of distinct spatial distribution of cortical atrophy and WM lesions predicting disability. The cerebellar involvement is shown as a key feature in the CNN framework for prediction of rapid clinical deterioration.

Collapse
Diagnostic Criteria and Differential Diagnosis Poster Presentation

P0263 - Serum neurofilament predicts clinical progression and increases diagnostic accuracy in patients with early multiple sclerosis (ID 1336)

Abstract

Background

Up to date prognostic estimation in newly diagnosed patients is hardly possible while the differentiation between disabling versus more benign courses is of utmost relevance. Reliable blood-based biomarkers that are associated with diagnosis and prognosis of multiple sclerosis (MS) have not been established.

Objectives

Can serum neurofilament light chain measurements serve as a reliable biomarker for diagnostic accuracy and prognosis for multiple sclerosis patients at the time point of diagnosis?

Methods

In a multicenter prospective longitudinal observational cohort, patients with a first diagnosis of multiple sclerosis (MS) or clinically isolated syndrome (CIS) were recruited between August 2010 and November 2015 in 22 centers and assessed yearly with a standardized protocol. Patients were offered standard immunotherapies according to national treatment guidelines. Serum NfL concentrations were measured using an ultrasensitive single-molecule array (Simoa).

Results

A possible association between sNfL levels and clinical diagnosis, relapses, MRI parameters and treatment decisions was tested in 814 patients classified according to current (2017) and older (2010) McDonald criteria at time point of diagnosis and two years after study inclusion sNfL levels correlated with number of T2 and Gd+ lesions and clinical relapses. After reclassification of CIS[2010] patients with existing CSF analysis, according to 2017 criteria, sNfL levels were lower in CIS[2017] than RRMS[2017] patients (9.1 pg/ml, IQR 6.2-13.7 pg/ml, n = 45; 10.8 pg/ml, IQR 7.4-20.1 pg/ml, n = 213; p = 0.036) and increased accuracy of distinction between CIS and RRMS, when including ≥ 90th percentile of sNfL values. Patients receiving disease-modifying treatment (DMT) during the first two years had higher sNfl baseline levels (11.8 pg/ml, 7.5-20.9 pg/ml, n = 727) than patients never receiving DMT (9.5 pg/ml, IQR 6.4-14.1 pg/ml, n = 87, p = 0.002). Longitudinal sNfL levels reflected treatment decisions within the first four years.

Conclusions

sNfL is associated with diagnosis and prognosis of MS patients at the time point of first diagnosis and may be of use for initial treatment stratification.

Collapse
Imaging Poster Presentation

P0599 - Linking microstructural integrity and motor cortex excitability in multiple sclerosis (ID 1151)

Speakers
Presentation Number
P0599
Presentation Topic
Imaging

Abstract

Background

Motor skills are commonly impaired in patients with multiple sclerosis (MS) as a consequence of gray (GM) and white matter (WM) pathology and cortical excitability abnormalities.

Objectives

We hypothesized that microstructural characteristics of motor regions as assessed with the neurite orientation dispersion and density imaging (NODDI) model predict motor cortical excitability that is frequently altered in MS. Further, we evaluated pathological microstructure alterations in motor WM tracts of MS patients compared to healthy controls (HC) using NODDI in comparison to the diffusion tensor imaging (DTI) parameter fractional anisotropy (FA).

Methods

We applied advanced diffusion imaging in 50 MS patients and 49 age-matched HC. As excitability maker, we assessed resting motor thresholds using non-invasive transcranial magnetic stimulation. For quantification of microstructural integrity of the motor system, neurite density index (NDI), orientation dispersion index (ODI), isotropic volume fraction (IVF) and FA averaged within left primary motor cortex as the stimulation site were considered. We applied hierarchical regression modeling to evaluate the prediction of the resting motor threshold by NDI, ODI, IVF and FA in MS patients and HC. Cognitive-motor performance quantified by the Nine Hole Peg Test and Trail Making Test part A (TMT-A) and part B (TMT-B) was regressed on the diffusion parameters in a subsample of 44 MS patients. In the WM, we applied tract-based spatial statistics with the threshold-free cluster enhancement (TFCE) method within motor tracts comparing MS patients and HC. We tracked contributions of NDI and ODI to FA and evaluated if the NODDI model detects additional pathological alterations.

Results

A hierarchical regression revealed that lower NDI suggestive for axonal loss in the GM significantly predicted higher motor thresholds, i.e. reduced excitability in MS patients (F(1,48) = 7.493, p = .009). Lower NDI was indicative for decreased performance in TMT-A (F(1,42) = 8.102; p = .007) and TMT-B (F(1,42) = 7.390; p = .009). Microstructural abnormalities of the interconnected WM tracts were characterized by lowered FA, decreased NDI and increased ODI in MS (all TFCE-corrected p < .05). NDI exclusively (56%) and in overlap with FA (19%) accounted for the largest amount of differences, followed by ODI alone (9%).

Conclusions

Our work shows that lower neurite density in primary motor cortex is linked to decreased motor cortical excitability and decreased cognitive-motor performance in MS patients. Lower neurite density and higher orientation dispersion are characteristic in the WM of MS patients compared to HC. Our results suggest that these markers are more sensitive to pathological alterations than the classical DTI measure FA. This work outlines the potential of microstructure imaging using advanced biophysical models to forecast neurodegeneration and excitability alterations in neuroinflammation.

Collapse
Imaging Poster Presentation

P0601 - Longitudinal functional modularisation and causality dynamics during de- and remyelination (ID 1715)

Abstract

Background

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system (CNS), one of its pathophysiological hallmarks is demyelination, which is known to be involved in neurodegenerative mechanisms.

Objectives

Modular architecture and its dynamic adaptation could play a critical role in achieving flexible alterations of cerebral network architecture during de- and remyelination which is still not fully elucidated.

Methods

We address dynamic adaptation to cuprizone model of general de- and remyelination and ask if network community organization can relate to the longitudinal time events. To start with baseline and then by introducing cuprizone into the diet of mice we induced full CNS demyelination by targeting oligodendrocytes, over a period of 5 weeks (two time points). A subsequent myelin synthesis was allowed over reintroduction of normal food (two time points). To identify the modular organization the resting state fMRI within the graph theory framework was analyzed from each of the five time points. The dynamic network reconfiguration was estimated by flexibility as parameter of modularity allegiance and effective connectivity analyses were applied to test the causality of network dynamics between the identified modules.

Results

We found six modules namely default mode network (DMN), hippocampus, thalamus, lateral cortical network, basal forebrain and ventral mid brain. Interestingly the dynamics of de- and remyelination was mirrored by an initial significant increase in flexibility values and a return to baseline in the hippocampus (F(4, 80) = 22.8, p < 0.001), DMN (F(4, 80) = 36.5, p < 0.001) and thalamus (F(4, 80) = 24.5, p < 0.001). The other three networks showed a reversed pattern. The strength of connections from the hippocampus to DMN was associated with the behavioral indicators of memory novel object recognition (NOR) (r2 = 0.3854, p < 0.001) and thalamus to hippocampus to locomotor activity (r2 = 0.3144, p < 0.001).

Conclusions

Taken together, our fMRI modular analyses showed that global modularity and flexibility partially compensate for demyelination. Dynamics of compensation could be identified as modular specific (i.e. hippocampus, thalamus and DMN) at different intermediate time points, supporting the hypothesis that altered thalamocortical connectivity is an early pathological hallmark of the disease. Causality dynamics also provide biomarkers for evaluating the course of MS and disease dynamics.

Collapse
Gender Differences, Hormones and Sex Chromosomes Poster Presentation

P1123 - Gender dimorphism of hippocampal intrinsic networks and regional integrity in multiple sclerosis (ID 1743)

Speakers
Presentation Number
P1123
Presentation Topic
Gender Differences, Hormones and Sex Chromosomes

Abstract

Background

The hippocampus is a complex anatomical structure with a fine-tuned intrinsic network architecture, shaped by functional and structural compartmentalization. The hippocampus is affected early in multiple sclerosis (MS) and besides focal neuroinflammatory damage, network disruption is thought to account for cognitive deficits in MS. Given the sex-related vulnerability to cognitive decline in MS, sex-driven differences in hippocampal networks and regional integrity can be hypothesized.

Objectives

To characterize sex effects on hippocampal network organization and subfield integrity, and their relation to cognitive performance.

Methods

In a cohort of 476 MS patients (age 35±10 years), 337 females and 139 males with a disease duration of 16±14 months were imaged on a 3T MRI scanner at baseline and after 2 years. A control group of healthy subjects (HS, n=110, age 34±15 years, 54 females) was included. Volumes of 12 hippocampal subfields were quantified and fed into the reconstruction of the single-subject morphometric networks and analyzed within the graph theoretical framework. Sex-related differences in network and subfield properties were evaluated with linear mixed-effects models, adjusted for age, center and total hippocampal volume; p-values are reported after Bonferroni correction for multiple comparisons.

Results

At baseline, both female and male patients displayed higher clustering (p<0.05) compared to HS. Female patients had higher clustering (p<0.05) but equally efficient network organization (local and global efficiency, p>0.05) compared to male patients. At follow-ups, independently of sex, patients had increased modularity, clustering and global efficiency, however, with higher values in female patients (all p<0.05). Both female and male patients had lower volumes in almost all subfields compared to HS. Female patients had smaller parasubiculum and presubiculum but larger molecular layer as compared to male patients. Over time, female patients had more widespread regional volumetric reduction compared to male patients. Cognitive performance was positively associated with clustering (r=0.27, p<0.01), local (r=0.25, p<0.01) and global efficiency (r=0.24, p<0.01) only in female but not in male patients.

Conclusions

Our findings suggest a more clustered and modular network architecture in female patients despite a more extensive local atrophy over time. The stronger association of cognitive performance with intrinsic hippocampal connectivity may explain cognitive reserve in female patients. These results may serve for sex-targeted neuropsychological interventions.

Collapse

Presenter Of 2 Presentations

Machine Learning/Network Science Poster Presentation

P0004 - Convolutional neural network framework for predicting progression in early MS (ID 1679)

Speakers
Presentation Number
P0004
Presentation Topic
Machine Learning/Network Science

Abstract

Background

Brain tissue damage is closely linked to disability in multiple sclerosis (MS). The localization of white matter (WM) lesions influences the course of the disease.

Objectives

However, the interrelation between lesions topography and cortical atrophy distribution for predicting the clinical disability remains unclear. Use a deep learning neural network framework with the purpose to identify critical co-varying patterns for individualized disease prediction.

Methods

Clinical disability was measured using the Expanded Disability Status Score at baseline and at a one-year follow-up in a cohort of 119 patients with early relapsing-remitting MS and in a replication cohort of 81 patients. Co-varying patterns of cortical atrophy and baseline lesion distribution were extracted by parallel ICA and used as features for constructing a deep learning convolutional neural network. The prediction was conducted for each identified lesion pattern separately using 50% as training cohort and 50% as testing cohort.

Results

In the study cohort, we identified three distinct distribution types of WM lesions (“cerebellar”, “bihemispheric” and “left-lateralized”). The “cerebellar” and “left-lateralized” patterns were reproducibly detected in the second cohort. Each of the patterns predicted to different extents, short-term disability progression, while the “cerebellar” pattern predicting individual disability progression with an 10-fold cross-validation accuracy of above 90% for the Study cohort (95% CI: 88%-94%) and above 85% for the replication cohort (95% CI: 81%-88%) respectively.

Conclusions

These findings highlight that role of distinct spatial distribution of cortical atrophy and WM lesions predicting disability. The cerebellar involvement is shown as a key feature in the CNN framework for prediction of rapid clinical deterioration.

Collapse
Imaging Poster Presentation

P0601 - Longitudinal functional modularisation and causality dynamics during de- and remyelination (ID 1715)

Abstract

Background

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system (CNS), one of its pathophysiological hallmarks is demyelination, which is known to be involved in neurodegenerative mechanisms.

Objectives

Modular architecture and its dynamic adaptation could play a critical role in achieving flexible alterations of cerebral network architecture during de- and remyelination which is still not fully elucidated.

Methods

We address dynamic adaptation to cuprizone model of general de- and remyelination and ask if network community organization can relate to the longitudinal time events. To start with baseline and then by introducing cuprizone into the diet of mice we induced full CNS demyelination by targeting oligodendrocytes, over a period of 5 weeks (two time points). A subsequent myelin synthesis was allowed over reintroduction of normal food (two time points). To identify the modular organization the resting state fMRI within the graph theory framework was analyzed from each of the five time points. The dynamic network reconfiguration was estimated by flexibility as parameter of modularity allegiance and effective connectivity analyses were applied to test the causality of network dynamics between the identified modules.

Results

We found six modules namely default mode network (DMN), hippocampus, thalamus, lateral cortical network, basal forebrain and ventral mid brain. Interestingly the dynamics of de- and remyelination was mirrored by an initial significant increase in flexibility values and a return to baseline in the hippocampus (F(4, 80) = 22.8, p < 0.001), DMN (F(4, 80) = 36.5, p < 0.001) and thalamus (F(4, 80) = 24.5, p < 0.001). The other three networks showed a reversed pattern. The strength of connections from the hippocampus to DMN was associated with the behavioral indicators of memory novel object recognition (NOR) (r2 = 0.3854, p < 0.001) and thalamus to hippocampus to locomotor activity (r2 = 0.3144, p < 0.001).

Conclusions

Taken together, our fMRI modular analyses showed that global modularity and flexibility partially compensate for demyelination. Dynamics of compensation could be identified as modular specific (i.e. hippocampus, thalamus and DMN) at different intermediate time points, supporting the hypothesis that altered thalamocortical connectivity is an early pathological hallmark of the disease. Causality dynamics also provide biomarkers for evaluating the course of MS and disease dynamics.

Collapse