Contributions
Abstract: P1032
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - Imaging
Introduction: Lesion volume measures and their changes over time are meaningful biomarkers in multiple sclerosis (MS) prognosis. Manual lesion segmentation for computing volume in a single or multiple time points is time consuming and suffers from intra and inter-observer variability. In this work, we present a fully-automatic longitudinal assessment of lesion changes from follow-up MRI scans. This has been integrated in the MSmetrix software, which is approved for clinical use in EU, Canada and Australia.
Methods: The presented approach is an iterative and unsupervised automatic method for longitudinal white matter lesion segmentation, which is based on a joint expectation-maximisation (EM) framework for two time points. 3D T1-weighted and 3D FLAIR MR images are used as input and lesions are segmented in three steps: (1) cross-sectional analysis, providing a prior on joint lesion segmentation of the two time points; (2) creation of difference image, which is used to model the lesion evolution as a Gaussian mixture model; (3) a joint EM lesion segmentation framework that takes as input the cross-sectional lesion segmentation and difference image to provide the final lesion segmentation at each time point. The accuracy and reproducibility of the software is evaluated.
Results: The accuracy of our proposed method, as evaluated by a median Dice score with a ground truth expert labeling data set, was 0.63 and the Pearson correlation coefficient was equal to 0.96. A test-retest data set was used to evaluate the reproducibility. The median absolute volume difference was 0.11 ml. Our presented method outperforms cross-sectional lesion segmentation methods and the longitudinal approach of the Lesion Segmentation Toolbox (LST).
Conclusion: Our results demonstrate that the proposed method has a high accuracy compared to expert manual labeling and a favourable reproducibility. This allows the use of lesion change measures in a clinical routine setting.
Disclosure: Saurabh Jain: employee of icometrix
Annemie Ribbens: employee of icometrix
Diana M. Sima: employee of icometrix
Melissa Cambron: nothing to disclose
Jacques De Keyser: nothing to disclose
Chenyu Wang: nothing to disclose
Michael H Barnett: nothing to disclose
Sabine Van Huffel: nothing to disclose
Frederik Maes: minority shareholder of icometrix
Anke Maertens: employee of icometrix
Dirk Smeets: employee of icometrix
Abstract: P1032
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - Imaging
Introduction: Lesion volume measures and their changes over time are meaningful biomarkers in multiple sclerosis (MS) prognosis. Manual lesion segmentation for computing volume in a single or multiple time points is time consuming and suffers from intra and inter-observer variability. In this work, we present a fully-automatic longitudinal assessment of lesion changes from follow-up MRI scans. This has been integrated in the MSmetrix software, which is approved for clinical use in EU, Canada and Australia.
Methods: The presented approach is an iterative and unsupervised automatic method for longitudinal white matter lesion segmentation, which is based on a joint expectation-maximisation (EM) framework for two time points. 3D T1-weighted and 3D FLAIR MR images are used as input and lesions are segmented in three steps: (1) cross-sectional analysis, providing a prior on joint lesion segmentation of the two time points; (2) creation of difference image, which is used to model the lesion evolution as a Gaussian mixture model; (3) a joint EM lesion segmentation framework that takes as input the cross-sectional lesion segmentation and difference image to provide the final lesion segmentation at each time point. The accuracy and reproducibility of the software is evaluated.
Results: The accuracy of our proposed method, as evaluated by a median Dice score with a ground truth expert labeling data set, was 0.63 and the Pearson correlation coefficient was equal to 0.96. A test-retest data set was used to evaluate the reproducibility. The median absolute volume difference was 0.11 ml. Our presented method outperforms cross-sectional lesion segmentation methods and the longitudinal approach of the Lesion Segmentation Toolbox (LST).
Conclusion: Our results demonstrate that the proposed method has a high accuracy compared to expert manual labeling and a favourable reproducibility. This allows the use of lesion change measures in a clinical routine setting.
Disclosure: Saurabh Jain: employee of icometrix
Annemie Ribbens: employee of icometrix
Diana M. Sima: employee of icometrix
Melissa Cambron: nothing to disclose
Jacques De Keyser: nothing to disclose
Chenyu Wang: nothing to disclose
Michael H Barnett: nothing to disclose
Sabine Van Huffel: nothing to disclose
Frederik Maes: minority shareholder of icometrix
Anke Maertens: employee of icometrix
Dirk Smeets: employee of icometrix