Author Of 1 Presentation
FC03.02 - A step forward toward the fully automated assessment of the central vein sign
Abstract
Background
A deep-learning prototype method, called CVSNet, was recently introduced for the automated detection of the central vein sign (CVS) in brain lesions and demonstrated effective and accurate discrimination of multiple sclerosis (MS) from its mimics. However, this method solely considered focal lesions displaying the central vein sign (CVS+) or not (CVS−), therefore requiring a manual pre-selection of the lesions to be evaluated by eliminating the so-called excluded lesions (CVSe) as defined by the NAIMS criteria. CVSe lesions may however play an important role in differential diagnosis. Moreover, extending the automated CVS classification to these lesions would facilitate the integration of CVSNet with existing MS lesion segmentation algorithms in a fully automated pipeline.
Objectives
To develop an improved version of the CVSNet prototype method able to classify all types of lesions (CVS+, CVS− and CVSe).
Methods
Patients with an established MS or CIS diagnosis (RRMS 29; SPMS 10; PPMS 10; CIS 1; mean ± SD age: 50 ± 11 years; male/female: 23/27), and healthy controls (n=8; mean ± SD age: 41 ± 9 years; male/female: 5/3), underwent 3T brain MRI (MAGNETOM Skyra and MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany, or Achieva, Philips Healthcare, Best, Netherlands). Brain lesions were automatically segmented and manually corrected by a single rater. CVS assessment was conducted on FLAIR* images by two raters, according to the NAIMS guidelines, yielding 1542 CVS+, 1004 CVS−, and 1131 CVSe lesions. A convolutional neural network (CNN) based on the CVSnet architecture was trained with different configurations using 3021 samples (1261 CVS+, 847 CVS−, and 913 CVSe) and evaluated in 656 unseen samples (281 CVS+, 157 CVS−, and 218 CVSe, from 13 patients) for final testing. The configurations relied on different combinations of the following channels as input: (i) FLAIR*, (ii) T2*, (iii) lesion mask, and (iv) CSF and brain tissue concentration maps obtained from a partial-volume estimation algorithm. Lesion-wise classification performance was evaluated for the different configurations by estimating the sensitivity, specificity, and accuracy for each lesion class.
Results
The results were similar across the different configurations. The best performance in the unseen testing set was obtained when all channels were used as input (sensitivity: 0.71, 0.73; specificity: 0.71, 0.81; and accuracy: 0.71, 0.79 for CVS+, CVS−, respectively). For CVSe, this approach achieved 0.52 sensitivity, 0.94 specificity, and 0.80 accuracy.
Conclusions
We introduced a modified CVSNet prototype method that can analyze the presence of the central vein for all types of brain lesions, enabling its integration with current MS lesion segmentation algorithms. This new feature will allow a fully automated assessment of the CVS in patients’ brains, speeding up the evaluation of CVS as a diagnostic biomarker for differentiating MS from mimicking diseases.
Presenter Of 1 Presentation
FC03.02 - A step forward toward the fully automated assessment of the central vein sign
Abstract
Background
A deep-learning prototype method, called CVSNet, was recently introduced for the automated detection of the central vein sign (CVS) in brain lesions and demonstrated effective and accurate discrimination of multiple sclerosis (MS) from its mimics. However, this method solely considered focal lesions displaying the central vein sign (CVS+) or not (CVS−), therefore requiring a manual pre-selection of the lesions to be evaluated by eliminating the so-called excluded lesions (CVSe) as defined by the NAIMS criteria. CVSe lesions may however play an important role in differential diagnosis. Moreover, extending the automated CVS classification to these lesions would facilitate the integration of CVSNet with existing MS lesion segmentation algorithms in a fully automated pipeline.
Objectives
To develop an improved version of the CVSNet prototype method able to classify all types of lesions (CVS+, CVS− and CVSe).
Methods
Patients with an established MS or CIS diagnosis (RRMS 29; SPMS 10; PPMS 10; CIS 1; mean ± SD age: 50 ± 11 years; male/female: 23/27), and healthy controls (n=8; mean ± SD age: 41 ± 9 years; male/female: 5/3), underwent 3T brain MRI (MAGNETOM Skyra and MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany, or Achieva, Philips Healthcare, Best, Netherlands). Brain lesions were automatically segmented and manually corrected by a single rater. CVS assessment was conducted on FLAIR* images by two raters, according to the NAIMS guidelines, yielding 1542 CVS+, 1004 CVS−, and 1131 CVSe lesions. A convolutional neural network (CNN) based on the CVSnet architecture was trained with different configurations using 3021 samples (1261 CVS+, 847 CVS−, and 913 CVSe) and evaluated in 656 unseen samples (281 CVS+, 157 CVS−, and 218 CVSe, from 13 patients) for final testing. The configurations relied on different combinations of the following channels as input: (i) FLAIR*, (ii) T2*, (iii) lesion mask, and (iv) CSF and brain tissue concentration maps obtained from a partial-volume estimation algorithm. Lesion-wise classification performance was evaluated for the different configurations by estimating the sensitivity, specificity, and accuracy for each lesion class.
Results
The results were similar across the different configurations. The best performance in the unseen testing set was obtained when all channels were used as input (sensitivity: 0.71, 0.73; specificity: 0.71, 0.81; and accuracy: 0.71, 0.79 for CVS+, CVS−, respectively). For CVSe, this approach achieved 0.52 sensitivity, 0.94 specificity, and 0.80 accuracy.
Conclusions
We introduced a modified CVSNet prototype method that can analyze the presence of the central vein for all types of brain lesions, enabling its integration with current MS lesion segmentation algorithms. This new feature will allow a fully automated assessment of the CVS in patients’ brains, speeding up the evaluation of CVS as a diagnostic biomarker for differentiating MS from mimicking diseases.
Author Of 2 Presentations
P0592 - In-clinic performance of MSPie, an image analysis prototype for automated MRI quantitative point-of-care metrics in MS (ID 1310)
Abstract
Background
Automated and reproducible measures of MS severity and subclinical inflammatory activity and neurodegeneration in routine practice could support therapeutic decisions and accelerate research. Toward this goal, we developed and validated a software prototype, MSPie (MS PATHS Image Evaluation). MSPie runs on syngo.via Frontier (Siemens Healthcare, Erlangen, Germany) and processes standardized T2 FLAIR and T1-weighted MRIs to quantify brain parenchymal fraction (BPF), T2 lesion volume, and #new/enlarging T2 lesions (NET2L). Results are reviewable by radiologists through an interface that displays current, prior, and subtraction images, as well as overlays of brain and lesion segmentations, and allows +/- corrections of NET2L.
Objectives
To assess an image analysis prototype integrated into radiological practice to generate quantitative brain volume and lesion measurements at the point of care.
Methods
MSPie was installed at 2 MS Partners Advancing Technology and Health Solutions (MS PATHS) institutions. 3 neuroradiologists per institution used MSPie to review 40 longitudinal pairs of routine MS PATHS MRIs. For each case, radiologists performed a visual assessment of the brain segmentation used for BPF, manually corrected NET2L if needed, approved or rejected the results, and completed a performance evaluation survey.
Results
MSPie performance was assessed in 240 cases. Radiologists accepted MSPie-generated BPF and lesion results for 230/240 cases (96%). 38.8% of cases required corrections of false positive (FP) or false negative (FN) NET2L, with a mean of 2.5 (FP+FN) NET2L per case. In 94% of cases, NET2L FP+FN was £3, a prespecified design target based on radiologists’ input. MSPie detected 221/229 true NET2L, yielding a sensitivity of 96.2%. In 18% of cases, radiologists reported MSPie-detected NET2L they would have missed. Mean performance ratings on a scale of 1(poor) to 5(excellent) were: 3.9 for overall performance; 3.9 for brain segmentation; 3.9 for T2 lesion segmentation.
Conclusions
Incorporation of brain volume and T2 lesion quantification into MS imaging practice is feasible. MSPie demonstrated a high sensitivity for disease activity, detecting some NET2L that might have been missed by radiologists. MSPie achieved the prespecified target rate of acceptable false positive NET2L. MSPie might allow neuroradiologists to provide quantitative brain atrophy and T2 lesion metrics in clinical practice and to increase their diagnostic precision.
Disclosures: MS PATHS is sponsored by Biogen.
P0594 - Interpreting brain parenchymal fraction by comparison to healthy volunteers: Initial results from the MS PATHS normative sub-study (ID 1646)
Abstract
Background
Mean rates of brain atrophy in healthy controls range from 0.05-0.5%, depending on age and technical factors, including scanner, acquisition sequences, and image analysis techniques. In MS PATHS (MS Partners Advancing Technology for Health Solutions), standardized MRIs are analyzed using a software prototype (MSPie, MS PATHS Image Evaluation) that incorporates a novel approach to calculate BPF. Normative ranges measured using MSPie are needed to distinguish age- and disease-related changes.
Objectives
To establish a normative reference for interpretation of brain parenchymal fraction (BPF) in individual MS patients relative to age-matched healthy volunteers (HV).
Methods
HV aged 21-60 were recruited at 6 MS PATHS sites to be age-, race-, and gender-matched to the MS PATHS cohort. HVs were imaged at baseline and once/year using 3T scanners (Siemens Healthcare, Erlangen, Germany) and standardized acquisitions (3DFLAIR and 3DT1), as in routine MRIs in MS PATHS. MRIs from UK Biobank supplemented the normative dataset past the age of 60. All MRIs were analyzed with MSPie to calculate BPF. BPF normative percentile were calculated for each age using quantile regression. Mean annualized rate of brain atrophy was estimated from HVs with follow-up MRIs. BPF percentiles were applied to the MS PATHS cohort. Mean Processing Speed Test (PST) z-scores were compared in MS patients stratified based on BPF percentiles.
Results
209 HVs were enrolled, 590 UKBiobank HVs were selected, and 9479 MS patients had at least one MRI. HV BPF values ranged from 0.855-0.895 in the 21-30 age group to 0.796-0.882 in the 61-73 age group, demonstrating accelerating and more variable atrophy with increasing age. For MS patients age 21-73 years (n=6791), mean age-adjusted BPF percentile was 27.8%, where BPF values fell above the 50th%-ile in 23.4% (“mild MS”) and below the 25th%-ile in 57.6% (“severe MS”). Mean PST z-scores differed in BPF-based mild MS vs. severe MS groups (-0.15 and -0.83; p<0.001). Mean annualized BPF change in HV was -0.08% (range: -0.71% to +0.57%) based on 71 subjects (mean age: 41.1 years) with >2 MRIs.
Conclusions
Incorporating normative reference data into MSPie will aid clinicians with interpretation of individual patients’ BPF in clinical practice and may enable patient stratification based on BPF and other predictors. Additional longitudinal normative data are being collected to contextualize disease progression as measured by BPF change over time.
Disclosures: MS PATHS is sponsored by Biogen.