Contributions
Abstract: P700
Type: Poster Sessions
Abstract Category: Clinical aspects of MS - Clinical assessment tools
Introduction: Lesion load (LL) in MS is a standard clinical trial outcome measure. However, the correlation between LL, as revealed by imaging, and symptoms is poor and the relationship with the spatial distribution of lesions remains obscure - thus the term 'radiological-clinical paradox'. We propose a new representation of MS lesions that can be extracted quickly and serve to narrow this causal gap.
Objectives: The objective of this work is to assess the usefulness of modeling information on lesion dissemination in space and their location on specific white matter tracts.
Methods: This retrospective study used data from definite MS patients, collected in 2011-2015 routine clinical visits at the Lausanne University Hospital Neuroimmunology/MS unit. We used a subset with both a 3D fluid-attenuated inversion recovery (FLAIR) sequence and manual lesion segmentations, in accordance with the institution's rules. Subjects were acquired on several scanners (one 1.5T and four 3T, all Siemens Healthcare, Erlangen, Germany). 77 subjects were included (23 males, median age 36.6, median EDSS 2 [range 0-8.5]).
In addition to lesion volume and LL, we used a 20-tracts probabilistic white matter atlas (JHU) to automatically count lesion/tracts overlaps, and represent each subject's tract-specific lesion presence probability. To describe the brain-wide topological distribution of lesions, we calculated a distance graph, where nodes are lesion centers of mass, from which we computed a clustering coefficient and an average path length for each subject; these describe local lesion clustering and average of local between-lesion distances.
Using rank regression, we predicted EDSS from these spatial features.
Results: The baseline model with sex, age, lesion volume, and LL shows an imaging-EDSS rank correlation of 0.51 (R2 0.38). Adding tract annotations and graph topological features increases the rank correlation to 0.68 (R2 0.69), a significant improvement despite the loss of degrees of freedom (p=0.03).
Conclusions: We have proposed an approach to automatically annotate lesions, which can be used on clinical routine data, with no additional diffusion acquisition, to quickly enrich analysis of the links between imaging findings and clinical outcomes. The annotations may be a sensitive imaging marker for MS disease activity, while providing convenient visualization of the most impacted tracts, helping to build a link with specific functional systems and neurological deficits.
Disclosure: Jonas Richiardi: Salary paid in part by Siemens Healthcare AG Switzerland.
Claudia Bigoni : Salary paid by Siemens Healthcare AG Switzerland.
Mario Joao Fartaria : Salary paid by Siemens Healthcare AG Switzerland.
Pietro Maggi : nothing to disclose
Myriam Schluep : nothing to disclose
Renaud Du Pasquier : nothing to disclose
Patric Hagmann : nothing to disclose
Reto Meuli : nothing to disclose
Tobias Kober : Salary paid by Siemens Healthcare AG Switzerland.
Abstract: P700
Type: Poster Sessions
Abstract Category: Clinical aspects of MS - Clinical assessment tools
Introduction: Lesion load (LL) in MS is a standard clinical trial outcome measure. However, the correlation between LL, as revealed by imaging, and symptoms is poor and the relationship with the spatial distribution of lesions remains obscure - thus the term 'radiological-clinical paradox'. We propose a new representation of MS lesions that can be extracted quickly and serve to narrow this causal gap.
Objectives: The objective of this work is to assess the usefulness of modeling information on lesion dissemination in space and their location on specific white matter tracts.
Methods: This retrospective study used data from definite MS patients, collected in 2011-2015 routine clinical visits at the Lausanne University Hospital Neuroimmunology/MS unit. We used a subset with both a 3D fluid-attenuated inversion recovery (FLAIR) sequence and manual lesion segmentations, in accordance with the institution's rules. Subjects were acquired on several scanners (one 1.5T and four 3T, all Siemens Healthcare, Erlangen, Germany). 77 subjects were included (23 males, median age 36.6, median EDSS 2 [range 0-8.5]).
In addition to lesion volume and LL, we used a 20-tracts probabilistic white matter atlas (JHU) to automatically count lesion/tracts overlaps, and represent each subject's tract-specific lesion presence probability. To describe the brain-wide topological distribution of lesions, we calculated a distance graph, where nodes are lesion centers of mass, from which we computed a clustering coefficient and an average path length for each subject; these describe local lesion clustering and average of local between-lesion distances.
Using rank regression, we predicted EDSS from these spatial features.
Results: The baseline model with sex, age, lesion volume, and LL shows an imaging-EDSS rank correlation of 0.51 (R2 0.38). Adding tract annotations and graph topological features increases the rank correlation to 0.68 (R2 0.69), a significant improvement despite the loss of degrees of freedom (p=0.03).
Conclusions: We have proposed an approach to automatically annotate lesions, which can be used on clinical routine data, with no additional diffusion acquisition, to quickly enrich analysis of the links between imaging findings and clinical outcomes. The annotations may be a sensitive imaging marker for MS disease activity, while providing convenient visualization of the most impacted tracts, helping to build a link with specific functional systems and neurological deficits.
Disclosure: Jonas Richiardi: Salary paid in part by Siemens Healthcare AG Switzerland.
Claudia Bigoni : Salary paid by Siemens Healthcare AG Switzerland.
Mario Joao Fartaria : Salary paid by Siemens Healthcare AG Switzerland.
Pietro Maggi : nothing to disclose
Myriam Schluep : nothing to disclose
Renaud Du Pasquier : nothing to disclose
Patric Hagmann : nothing to disclose
Reto Meuli : nothing to disclose
Tobias Kober : Salary paid by Siemens Healthcare AG Switzerland.