Contributions
Abstract: P576
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - 21 Imaging
Background: Lesion load is a commonly used biomarker of disease status in multiple sclerosis (MS), yet it has shown poor association with clinical outcomes. Additional information relevant to disease severity is thought to be captured in the natural history of lesion formation, but this feature is difficult to assess in patients with large confluent lesions. To address this issue, a novel statistical technique is introduced for identifying pathologically distinct lesions within regions of confluent lesion tissue.
Methods: Clinical and magnetic resonance imaging (MRI) data were obtained from 60 subjects with MS as part of a natural history study at the NINDS. T1-weighted and T2-weighted sequences were acquired at monthly visits. To obtain maps of distinct lesions, a multi-stage algorithm was used to segment lesion tissue, find Hessian-based lesion centers, and cluster lesion tissue around each center. The lesion count given by this technique is henceforth referred to as Cn.
Validation: Lesion counts obtained by the novel method (Cn) and counts obtained by a basic “connected components” technique (Cb) were compared to 'gold standard' temporally-informed counts (Cg) for the subset of lesions that appeared while subjects were in the study. The correlation between Cn and Cg was greater than the correlation between Cb and Cg (rng = .89, rbg = .69,
p = .002). The difference between Cn and Cg was not significant (t59 = 1.56, p = .12), while the difference between Cb and Cg was highly significant (t59 = 4.40, p < .001). Together, these findings suggest that the novel technique is a significant improvement over the basic method, and provides valid lesion counts.
Results: Accounting for lesion load and age, Cn was negatively associated with EDSS (t58 = -2.86,
p = .006), suggesting that for a given lesion load, a higher lesion count (thus a lower average size) is associated with lower disease severity. The inclusion of Cn in the model explains an additional 10% of the variance in EDSS, providing support to the idea that lesion count contains disease information independent of lesion load.
Conclusion: This study introduces a technique for separating spatially connected lesion tissue into pathologically distinct lesion components, and shows it to be both valid and clinically relevant. These findings demonstrate that it is possible to recover the natural history of MS lesion formation in an automated fashion using MRI scans from a single cross-sectional visit.
Disclosure:
Jordan Dworkin: Nothing to disclose
Ipek Oguz: Nothing to disclose
Kristin Linn: Nothing to disclose
Greg Fleischman: Nothing to disclose
Paul Yushkevich: Nothing to disclose
Daniel Reich: Research Funding from Vertex Pharmaceuticals and the Myelin Repair Foundation
Russell Shinohara: Consulting and advisory boards for Roche and Genentech
Abstract: P576
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - 21 Imaging
Background: Lesion load is a commonly used biomarker of disease status in multiple sclerosis (MS), yet it has shown poor association with clinical outcomes. Additional information relevant to disease severity is thought to be captured in the natural history of lesion formation, but this feature is difficult to assess in patients with large confluent lesions. To address this issue, a novel statistical technique is introduced for identifying pathologically distinct lesions within regions of confluent lesion tissue.
Methods: Clinical and magnetic resonance imaging (MRI) data were obtained from 60 subjects with MS as part of a natural history study at the NINDS. T1-weighted and T2-weighted sequences were acquired at monthly visits. To obtain maps of distinct lesions, a multi-stage algorithm was used to segment lesion tissue, find Hessian-based lesion centers, and cluster lesion tissue around each center. The lesion count given by this technique is henceforth referred to as Cn.
Validation: Lesion counts obtained by the novel method (Cn) and counts obtained by a basic “connected components” technique (Cb) were compared to 'gold standard' temporally-informed counts (Cg) for the subset of lesions that appeared while subjects were in the study. The correlation between Cn and Cg was greater than the correlation between Cb and Cg (rng = .89, rbg = .69,
p = .002). The difference between Cn and Cg was not significant (t59 = 1.56, p = .12), while the difference between Cb and Cg was highly significant (t59 = 4.40, p < .001). Together, these findings suggest that the novel technique is a significant improvement over the basic method, and provides valid lesion counts.
Results: Accounting for lesion load and age, Cn was negatively associated with EDSS (t58 = -2.86,
p = .006), suggesting that for a given lesion load, a higher lesion count (thus a lower average size) is associated with lower disease severity. The inclusion of Cn in the model explains an additional 10% of the variance in EDSS, providing support to the idea that lesion count contains disease information independent of lesion load.
Conclusion: This study introduces a technique for separating spatially connected lesion tissue into pathologically distinct lesion components, and shows it to be both valid and clinically relevant. These findings demonstrate that it is possible to recover the natural history of MS lesion formation in an automated fashion using MRI scans from a single cross-sectional visit.
Disclosure:
Jordan Dworkin: Nothing to disclose
Ipek Oguz: Nothing to disclose
Kristin Linn: Nothing to disclose
Greg Fleischman: Nothing to disclose
Paul Yushkevich: Nothing to disclose
Daniel Reich: Research Funding from Vertex Pharmaceuticals and the Myelin Repair Foundation
Russell Shinohara: Consulting and advisory boards for Roche and Genentech