•Uncertainty enables deep learning frameworks to be adopted into clinical workflows.•Uncertainty reflects challenge of small lesion detection, lesion contour segmentation.•MC dropout uncertainty ...measures correlate with incorrect predictions.
Display omitted
Deep learning networks have recently been shown to outperform other segmentation methods on various public, medical-image challenge datasets, particularly on metrics focused on large pathologies. For diseases such as Multiple Sclerosis (MS), however, monitoring all the focal lesions visible on MRI sequences, even very small ones, is essential for disease staging, prognosis, and evaluating treatment efficacy. Small lesion segmentation presents significant challenges to popular deep learning models. This, coupled with their deterministic predictions, hinders their clinical adoption. Uncertainty estimates for these predictions would permit subsequent revision by clinicians. We present the first exploration of multiple uncertainty estimates based on Monte Carlo (MC) dropout (Gal and Ghahramani, 2016) in the context of deep networks for lesion detection and segmentation in medical images. Specifically, we develop a 3D MS lesion segmentation CNN, augmented to provide four different voxel-based uncertainty measures based on MC dropout. We train the network on a proprietary, large-scale, multi-site, multi-scanner, clinical MS dataset, and compute lesion-wise uncertainties by accumulating evidence from voxel-wise uncertainties within detected lesions. We analyze the performance of voxel-based segmentation and lesion-level detection by choosing operating points based on the uncertainty. Uncertainty filtering improves both voxel and lesion-wise TPR and FDR on remaining, certain predictions compared to sigmoid-based TPR/FDR curves. Small lesions and lesion-boundaries are the most uncertain regions, which is consistent with human-rater variability.
IMPORTANCE: Accumulation of disability in multiple sclerosis may occur as relapse-associated worsening (RAW) or steady progression independent of relapse activity (PIRA), with PIRA regarded as a ...feature of primary and secondary progressive multiple sclerosis. OBJECTIVE: To investigate the contributions of relapse-associated worsening vs relapse-independent progression to overall confirmed disability accumulation (CDA) and assess respective baseline prognostic factors and outcomes of 2 treatments. DESIGN, SETTING, AND PARTICIPANTS: Analyses occurred from July 2015 to February 2020 on pooled data from the intention-to-treat population of 2 identical, phase 3, multicenter, double-blind, double-dummy, parallel-group randomized clinical trials (OPERA I and II) conducted between August 2011 and April 2015. In the trials, patients with relapsing multiple sclerosis (RMS), diagnosed using the 2010 revised McDonald criteria, were randomized from 307 trial sites in 56 countries; resulting data were analyzed in the pooled data set. INTERVENTIONS: Participants were randomized 1:1 to receive 600 mg of ocrelizumab by intravenous infusion every 24 weeks or subcutaneous interferon β-1a 3 times a week at a dose of 44 μg throughout a 96-week treatment period. MAIN OUTCOMES AND MEASURES: Confirmed disability accumulation was defined by an increase in 1 or more of 3 measures (Expanded Disability Status Scale, timed 25-ft walk, or 9-hole peg test), confirmed after 3 or 6 months, and classified per temporal association with confirmed clinical relapses (PIRA or RAW). RESULTS: In the pooled OPERA I and II population (1656 of 2096 eligible participants), baseline demographics and disease characteristics were similar for patients randomized to interferon β-1a vs ocrelizumab (mean SD age, 37.2 9.2 vs 37.1 9.2 years; 552 66.6% vs 541 women 65.4%). After 96 weeks, 12-week composite CDA had occurred in 223 (29.6% by Kaplan-Meier estimate) randomized to interferon β-1a and 167 (21.1%) randomized to ocrelizumab; 24-week composite CDA had occurred in 170 (22.7%) taking interferon β-1a and 129 (16.2%) taking ocrelizumab. The PIRA events were the main contributors to 12-week and 24-week composite CDA after 96 weeks in patients treated with interferon β-1a (174 of 223 78.0% and 137 of 170 80.6%, respectively) and ocrelizumab (147 of 167 88.0% and 115 of 129 89.1%, respectively); a minority had CDA explained by RAW events (69 of 390 17.7% and 52 of 299 17.4%, respectively). Very few patients with composite CDA experienced both RAW and PIRA events (17 of 390 4.4% for 12-week and 15 of 299 5.0% for 24-week composite CDA). Ocrelizumab (vs interferon β-1a) was associated with reduced risk of composite CDA (hazard ratio HR, 0.67) and confirmed PIRA (HR, 0.78) and RAW (HR, 0.47) events. CONCLUSIONS AND RELEVANCE: Most disability accumulation in RMS is not associated with overt relapses. This indicates an underlying progression in this typical RMS population and challenges the current clinical distinction of relapsing and progressive forms of multiple sclerosis. Ocrelizumab was superior to interferon β-1a in preventing both RAW and PIRA. TRIAL REGISTRATION: ClinicalTrials.gov Identifiers: OPERA I (NCT01247324) and OPERA II (NCT01412333).
In a randomized trial, patients who received evobrutinib, an inhibitor of Bruton’s tyrosine kinase, at 75 mg daily had significantly fewer enhancing lesions on MRI during weeks 12 through 24 than ...those who received placebo. However, there was no significant between-group difference for either a lower or a higher dose of evobrutinib, or in the annualized relapse rate or disability progression at any dose.
Objective
To evaluate the extent to which treatment effect on brain atrophy is able to mediate, at the trial level, the treatment effect on disability progression in relapsing–remitting multiple ...sclerosis (RRMS).
Methods
We collected all published randomized clinical trials in RRMS lasting at least 2 years and including as endpoints disability progression (defined as 6 or 3 months confirmed 1‐point increase on the Expanded Disability Status Scale), active magnetic resonance imaging (MRI) lesions (defined as new/enlarging T2 lesions), and brain atrophy (defined as change in brain volume between month 24 and month 6–12). Treatment effects were expressed as relative reductions. A linear regression, weighted for trial size and duration, was used to assess the relationship between the treatment effects on MRI markers and on disability progression.
Results
Thirteen trials including >13,500 RRMS patients were included in the meta‐analysis. Treatment effects on disability progression were correlated with treatment effects both on brain atrophy (R2 = 0.48, p = 0.001) and on active MRI lesions (R2 = 0.61, p < 0.001). When the effects on both MRI endpoints were included in a multivariate model, the correlation was higher (R2 = 0.75, p < 0.001), and both variables were retained as independently related to the treatment effect on disability progression.
Interpretation
In RRMS, the treatment effect on brain atrophy is correlated with the effect on disability progression over 2 years. This effect is independent of the effect of active MRI lesions on disability; the 2 MRI measures predict the treatment effect on disability more closely when used in combination. ANN NEUROL 2014;75:43–49
Chronic active and slowly expanding lesions with smouldering inflammation are neuropathological correlates of progressive multiple sclerosis pathology. T1 hypointense volume and signal intensity on ...T1-weighted MRI reflect brain tissue damage that may develop within newly formed acute focal inflammatory lesions or in chronic pre-existing lesions without signs of acute inflammation. Using a recently developed method to identify slowly expanding/evolving lesions in vivo from longitudinal conventional T2- and T1-weighted brain MRI scans, we measured the relative amount of chronic lesion activity as measured by change in T1 volume and intensity within slowly expanding/evolving lesions and non-slowly expanding/evolving lesion areas of baseline pre-existing T2 lesions, and assessed the effect of ocrelizumab on this outcome in patients with primary progressive multiple sclerosis participating in the phase III, randomized, placebo-controlled, double-blind ORATORIO study (n = 732, NCT01194570). We also assessed the predictive value of T1-weighted measures of chronic lesion activity for clinical multiple sclerosis progression as reflected by a composite disability measure including the Expanded Disability Status Scale, Timed 25-Foot Walk and 9-Hole Peg Test. We observed in this clinical trial population that most of total brain non-enhancing T1 hypointense lesion volume accumulation was derived from chronic lesion activity within pre-existing T2 lesions rather than new T2 lesion formation. There was a larger decrease in mean normalized T1 signal intensity and greater relative accumulation of T1 hypointense volume in slowly expanding/evolving lesions compared with non-slowly expanding/evolving lesions. Chronic white matter lesion activity measured by longitudinal T1 hypointense lesion volume accumulation in slowly expanding/evolving lesions and in non-slowly expanding/evolving lesion areas of pre-existing lesions predicted subsequent composite disability progression with consistent trends on all components of the composite. In contrast, whole brain volume loss and acute lesion activity measured by longitudinal T1 hypointense lesion volume accumulation in new focal T2 lesions did not predict subsequent composite disability progression in this trial at the population level. Ocrelizumab reduced longitudinal measures of chronic lesion activity such as T1 hypointense lesion volume accumulation and mean normalized T1 signal intensity decrease both within regions of pre-existing T2 lesions identified as slowly expanding/evolving and in non-slowly expanding/evolving lesions. Using conventional brain MRI, T1-weighted intensity-based measures of chronic white matter lesion activity predict clinical progression in primary progressive multiple sclerosis and may qualify as a longitudinal in vivo neuroimaging correlate of smouldering demyelination and axonal loss in chronic active lesions due to CNS-resident inflammation and/or secondary neurodegeneration across the multiple sclerosis disease continuum.
In this randomized trial involving patients with multiple sclerosis, BG-12 (dimethyl fumarate) reduced clinical relapses, disability progression, and MRI lesions. BG-12 treatment resulted in reduced ...lymphocyte counts and elevated liver aminotransferase levels.
Oral BG-12 (dimethyl fumarate) is being investigated for the treatment of multiple sclerosis. Inflammation and oxidative stress are central pathologic factors in multiple sclerosis.
1
,
2
Immune cell activation and infiltration into the central nervous system are thought to result in widespread cellular damage, potentially owing to the dysregulated production and release of reactive oxygen and nitrogen species, such as hydrogen peroxide and peroxynitrite, and proinflammatory stimuli.
3
This combination of toxic factors ultimately results in demyelination and neurodegeneration, causing disease activity and progression of disability.
BG-12 has been shown to have beneficial effects in preclinical models of neuroinflammation, neurodegeneration, and toxic . . .
Multiple sclerosis (MS) can be divided into four phenotypes based on clinical evolution. The pathophysiological boundaries of these phenotypes are unclear, limiting treatment stratification. Machine ...learning can identify groups with similar features using multidimensional data. Here, to classify MS subtypes based on pathological features, we apply unsupervised machine learning to brain MRI scans acquired in previously published studies. We use a training dataset from 6322 MS patients to define MRI-based subtypes and an independent cohort of 3068 patients for validation. Based on the earliest abnormalities, we define MS subtypes as cortex-led, normal-appearing white matter-led, and lesion-led. People with the lesion-led subtype have the highest risk of confirmed disability progression (CDP) and the highest relapse rate. People with the lesion-led MS subtype show positive treatment response in selected clinical trials. Our findings suggest that MRI-based subtypes predict MS disability progression and response to treatment and may be used to define groups of patients in interventional trials.
Display omitted
► Systematic overview of the MS segmentation literature. ► Comprehensive description of methods. ► Description of MS lesions segmentation validation. ► No method deals completely with ...the full complexity of lesions.
Magnetic resonance (MR) imaging is often used to characterize and quantify multiple sclerosis (MS) lesions in the brain and spinal cord. The number and volume of lesions have been used to evaluate MS disease burden, to track the progression of the disease and to evaluate the effect of new pharmaceuticals in clinical trials. Accurate identification of MS lesions in MR images is extremely difficult due to variability in lesion location, size and shape in addition to anatomical variability between subjects. Since manual segmentation requires expert knowledge, is time consuming and is subject to intra- and inter-expert variability, many methods have been proposed to automatically segment lesions.
The objective of this study was to carry out a systematic review of the literature to evaluate the state of the art in automated multiple sclerosis lesion segmentation. From 1240hits found initially with PubMed and Google scholar, our selection criteria identified 80 papers that described an automatic lesion segmentation procedure applied to MS. Only 47 of these included quantitative validation with at least one realistic image. In this paper, we describe the complexity of lesion segmentation, classify the automatic MS lesion segmentation methods found, and review the validation methods applied in each of the papers reviewed. Although many segmentation solutions have been proposed, including some with promising results using MRI data obtained on small groups of patients, no single method is widely employed due to performance issues related to the high variability of MS lesion appearance and differences in image acquisition. The challenge remains to provide segmentation techniques that work in all cases regardless of the type of MS, duration of the disease, or MRI protocol, and this within a comprehensive, standardized validation framework. MS lesion segmentation remains an open problem.
Summary Background Subcutaneous pegylated interferon (peginterferon) beta-1a is being developed for treatment of relapsing multiple sclerosis, with less frequent dosing than currently available ...first-line injectable treatments. We assessed the safety and efficacy of peginterferon beta-1a after 48 weeks of treatment in the placebo-controlled phase of the ADVANCE trial, a study of patients with relapsing-remitting multiple sclerosis. Methods We did this 2-year, double-blind, parallel group, phase 3 study, with a placebo-controlled design for the first 48 weeks, at 183 sites in 26 countries. Patients with relapsing-remitting multiple sclerosis (age 18–65 years, with Expanded Disability Status Scale score ≤5) were randomly assigned (1:1:1) via an interactive voice response or web system, and stratified by site, to placebo or subcutaneous peginterferon beta-1a 125 μg once every 2 weeks or every 4 weeks. The primary endpoint was annualised relapse rate at 48 weeks. This trial is registered with ClinicalTrials.gov , number NCT00906399. Findings We screened 1936 patients and enrolled 1516, of whom 1512 were randomly assigned (500 to placebo, 512 to peginterferon every 2 weeks, 500 to peginterferon every 4 weeks); 1332 (88%) patients completed 48 weeks of treatment. Adjusted annualised relapse rates were 0·397 (95% CI 0·328–0·481) in the placebo group versus 0·256 (0·206–0·318) in the every 2 weeks group and 0·288 (0·234–0·355) in the every 4 weeks group (rate ratio for every 2 weeks group 0·644, 95% CI 0·500–0·831, p=0·0007; rate ratio for the every 4 weeks group 0·725, 95% CI 0·565–0·930, p=0·0114). 417 (83%) patients taking placebo, 481 (94%) patients taking peginterferon every 2 weeks, and 472 (94%) patients taking peginterferon every 4 weeks reported adverse events including relapses. The most common adverse events associated with peginterferon beta-1a were injection site reactions, influenza-like symptoms, pyrexia, and headache. 76 (15%) patients taking placebo, 55 (11%) patients taking study drug every 2 weeks, and 71 (14%) patients taking study drug every 4 weeks reported serious adverse events; relapse, pneumonia, and urinary tract infection were the most common. Interpretation After 48 weeks, peginterferon beta-1a significantly reduced relapse rate compared with placebo. The drug might be an effective treatment for relapsing-remitting multiple sclerosis with less frequent administration than available treatments. Funding Biogen Idec.
Summary Background Strong immunosuppression, including chemotherapy and immune-depleting antibodies followed by autologous haemopoietic stem-cell transplantation (aHSCT), has been used to treat ...patients with multiple sclerosis, improving control of relapsing disease. We addressed whether near-complete immunoablation followed by immune cell depleted aHSCT would result in long-term control of multiple sclerosis. Methods We did this phase 2 single-arm trial at three hospitals in Canada. We enrolled patients with multiple sclerosis, aged 18–50 years with poor prognosis, ongoing disease activity, and an Expanded Disability Status Scale of 3·0–6·0. Autologous CD34 selected haemopoietic stem-cell grafts were collected after mobilisation with cyclophosphamide and filgrastim. Immunoablation with busulfan, cyclophosphamide, and rabbit anti-thymocyte globulin was followed by aHSCT. The primary outcome was multiple sclerosis activity-free survival (events were clinical relapse, appearance of a new or Gd-enhancing lesion on MRI, and sustained progression of Expanded Disability Status Scale score). This study was registered at ClinicalTrials.gov , NCT01099930. Findings Between diagnosis and aHSCT, 24 patients had 167 clinical relapses over 140 patient-years with 188 Gd-enhancing lesions on 48 pre-aHSCT MRI scans. Median follow-up was 6·7 years (range 3·9–12·7). The primary outcome, multiple sclerosis activity-free survival at 3 years after transplantation was 69·6% (95% CI 46·6–84·2). With up to 13 years of follow-up after aHSCT, no relapses occurred and no Gd enhancing lesions or new T2 lesions were seen on 314 MRI sequential scans. The rate of brain atrophy decreased to that expected for healthy controls. One of 24 patients died of transplantation-related complications. 35% of patients had a sustained improvement in their Expanded Disability Status Scale score. Interpretation We describe the first treatment to fully halt all detectable CNS inflammatory activity in patients with multiple sclerosis for a prolonged period in the absence of any ongoing disease-modifying drugs. Furthermore, many of the patients had substantial recovery of neurological function despite their disease's aggressive nature. Funding Multiple Sclerosis Scientific Research Foundation.
PDF
