Longitudinal analysis of white matter and cortical lesions in multiple sclerosis

Meritxell Bach Cuadra, Mario Joao Fartaria de Oliveira, Cristina Granziera, Tobias Kober
2019
Article

Résumé

Purpose: The goals of this study were to assess the performance of a novel lesion segmentation tool for longitudinal analyses, as well as to validate the generated lesion progression map between two time points using conventional and non-conventional MR sequences. Material and methods: The lesion segmentation approach was evaluated with (LeMan-PV) and without (LeMan) the partial volume framework using "conventional" and "non-conventional" MR imaging in a two-year follow-up prospective study of 32 early RAMS patients. Manual segmentations of new, enlarged, shrunken, and stable lesions were used to evaluate the performance of the method variants. The true positive rate was estimated for those lesion evolutions in both white matter and cortex. The number of false positives was compared with two strategies for longitudinal analyses. New lesion tissue volume estimation was evaluated using Bland-Altman plots. Wilcoxon signed-rank test was used to evaluate the different setups. Results: The best median of the true positive rate was obtained using LeMan-PV with non-conventional sequences (P < .05): 87%, 87%, 100%, 83%, for new, enlarged, shrunken, and stable WM lesions, and 50%, 60%, 50%, 80%, for new, enlarged, shrunken, and stable cortical lesions, respectively. Most of the missed lesions were below the mean lesion size in each category. Lesion progression maps presented a median of 0 false positives (range:0-9) and the partial volume framework improved the volume estimation of new lesion tissue. Conclusion: LeMan-PV exhibited the best performance in the detection of new, enlarged, shrunken and stable WM lesions. The method showed lower performance in the detection of cortical lesions, likely due to their low occurrence, small size and low contrast with respect to surrounding tissues. The proposed lesion progression map might be useful in clinical trials or clinical routine.

Official source

https://infoscience.epfl.ch/record/270856?ln=fr

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Concepts associés

Chargement

Publications associées

Chargement

Meritxell Bach Cuadra, Mario Joao Fartaria de Oliveira, Cristina Granziera, Tobias Kober
2019
Article

Résumé

Official source

https://infoscience.epfl.ch/record/270856?ln=fr

À propos de ce résultat

Concepts associés (9)

Concepts associés

Chargement

Publications associées (7)

Chargement

Afficher plus

Publications associées

Chargement

Publications associées (7)

CVSnet: A machine learning approach for automated central vein sign assessment in multiple sclerosis

Meritxell Bach Cuadra, Mario Joao Fartaria de Oliveira, João Pedro Forjaco Jorge, Cristina Granziera, Tobias Kober, Francesco La Rosa, Reto Meuli, Jonas Richiardi

The central vein sign (CVS) is an efficient imaging biomarker for multiple sclerosis (MS) diagnosis, but its application in clinical routine is limited by inter-rater variability and the expenditure of time associated with manual assessment. We describe a deep learning-based prototype for automated assessment of the CVS in white matter MS lesions using data from three different imaging centers. We retrospectively analyzed data from 3 T magnetic resonance images acquired on four scanners from two different vendors, including adults with MS (n = 42), MS mimics (n = 33, encompassing 12 distinct neurological diseases mimicking MS) and uncertain diagnosis (n = 5). Brain white matter lesions were manually segmented on FLAIR* images. Perivenular assessment was performed according to consensus guidelines and used as ground truth, yielding 539 CVS-positive (CVS+) and 448 CVS-negative (CVS-) lesions. A 3D convolutional neural network ("CVSnet") was designed and trained on 47 datasets, keeping 33 for testing. FLAIR* lesion patches of CVS+/CVS- lesions were used for training and validation (n = 375/298) and for testing (n = 164/150). Performance was evaluated lesion-wise and subject-wise and compared with a state-of-the-art vesselness filtering approach through McNemar's test. The proposed CVSnet approached human performance, with lesion-wise median balanced accuracy of 81%, and subject-wise balanced accuracy of 89% on the validation set, and 91% on the test set. The process of CVS assessment, in previously manually segmented lesions, was similar to 600-fold faster using the proposed CVSnet compared with human visual assessment (test set: 4 seconds vs. 40 minutes). On the validation and test sets, the lesion-wise performance outperformed the vesselness filter method (P < 0.001). The proposed deep learning prototype shows promising performance in differentiating MS from its mimics. Our approach was evaluated using data from different hospitals, enabling larger multicenter trials to evaluate the benefit of introducing the CVS marker into MS diagnostic criteria.

2020

Chargement

, ,

Introduction. Development of the fetal brain surface with concomitant gyrification is one of the major maturational processes of the human brain. First delineated by postmortem studies or by ultrasound, MRI has recently become a powerful tool for studying in vivo the structural correlates of brain maturation. However, the quantitative measurement of fetal brain development is a major challenge because of the movement of the fetus inside the amniotic cavity, the poor spatial resolution, the partial volume effect and the changing appearance of the developing brain. Today extensive efforts are made to deal with the “post-acquisition” reconstruction of high-resolution 3D fetal volumes based on several acquisitions with lower resolution (Rousseau, F., 2006; Jiang, S., 2007). We here propose a framework devoted to the segmentation of the basal ganglia, the gray-white tissue segmentation, and in turn the 3D cortical reconstruction of the fetal brain. Method. Prenatal MR imaging was performed with a 1-T system (GE Medical Systems, Milwaukee) using single shot fast spin echo (ssFSE) sequences in fetuses aged from 29 to 32 gestational weeks (slice thickness 5.4mm, in plane spatial resolution 1.09mm). For each fetus, 6 axial volumes shifted by 1 mm were acquired (about 1 min per volume). First, each volume is manually segmented to extract fetal brain from surrounding fetal and maternal tissues. Inhomogeneity intensity correction and linear intensity normalization are then performed. A high spatial resolution image of isotropic voxel size of 1.09 mm is created for each fetus as previously published by others (Rousseau, F., 2006). B-splines are used for the scattered data interpolation (Lee, 1997). Then, basal ganglia segmentation is performed on this super reconstructed volume using active contour framework with a Level Set implementation (Bach Cuadra, M., 2010). Once basal ganglia are removed from the image, brain tissue segmentation is performed (Bach Cuadra, M., 2009). The resulting white matter image is then binarized and further given as an input in the Freesurfer software (http://surfer.nmr.mgh.harvard.edu/) to provide accurate three-dimensional reconstructions of the fetal brain. Results. High-resolution images of the cerebral fetal brain, as obtained from the low-resolution acquired MRI, are presented for 4 subjects of age ranging from 29 to 32 GA. An example is depicted in Figure 1. Accuracy in the automated basal ganglia segmentation is compared with manual segmentation using measurement of Dice similarity (DSI), with values above 0.7 considering to be a very good agreement. In our sample we observed DSI values between 0.785 and 0.856. We further show the results of gray-white matter segmentation overlaid on the high-resolution gray-scale images. The results are visually checked for accuracy using the same principles as commonly accepted in adult neuroimaging. Preliminary 3D cortical reconstructions of the fetal brain are shown in Figure 2. Conclusion. We hereby present a complete pipeline for the automated extraction of accurate three-dimensional cortical surface of the fetal brain. These results are preliminary but promising, with the ultimate goal to provide “movie” of the normal gyral development. In turn, a precise knowledge of the normal fetal brain development will allow the quantification of subtle and early but clinically relevant deviations. Moreover, a precise understanding of the gyral development process may help to build hypotheses to understand the pathogenesis of several neurodevelopmental conditions in which gyrification have been shown to be altered (e.g. schizophrenia, autism…). References. Rousseau, F. (2006), 'Registration-Based Approach for Reconstruction of High-Resolution In Utero Fetal MR Brain images', IEEE Transactions on Medical Imaging, vol. 13, no. 9, pp. 1072-1081. Jiang, S. (2007), 'MRI of Moving Subjects Using Multislice Snapshot Images With Volume Reconstruction (SVR): Application to Fetal, Neonatal, and Adult Brain Studies', IEEE Transactions on Medical Imaging, vol. 26, no. 7, pp. 967-980. Lee, S. (1997), 'Scattered data interpolation with multilevel B-splines', IEEE Transactions on Visualization and Computer Graphics, vol. 3, no. 3, pp. 228-244. Bach Cuadra, M. (2010), 'Central and Cortical Gray Mater Segmentation of Magnetic Resonance Images of the Fetal Brain', ISMRM Conference. Bach Cuadra, M. (2009), 'Brain tissue segmentation of fetal MR images', MICCAI.

2010

Chargement

Toward Reliable MR-Based Brain Structure Morphometry

Bénédicte Marie Maréchal

Dementing neurodegenerative disorders have more and more prominent public health implications with the most frequent cause being the currently non reversible Alzheimer's Disease (AD). However, over the past 15 years, an abundant literature has demonstrated the particular value of structural MRI from which brain morphological changes can be detected using automated volumetry and morphometry. These techniques offer promise to diagnose AD in early potentially-reversible stages as regional atrophy is currently "the" structural hallmark. Nevertheless, an essential question arises: to what extent are these quantitative measures dependent upon image quality? MRI quality can be mainly affected by machine-specific or patient-related artifacts, with motion being the most significant source of degradations. As we are currently not aware of any technique that reliably captures and corrects patient motion particularly for long high-resolution anatomical scans, rigorous quality assessment is hence of great importance to reliably derive qualitative or quantitative diagnostic information and consists in the first challenge addressed in this thesis. We developed a fully-automated method to classify the image quality based on a single MR image. Quality measures are derived by analyzing the air background of magnitude image and are capable of detecting image degradation from several sources, including bulk motion, residual magnetization from incomplete spoiling, blurring and ghosting. The method has been validated on 1670 multi-slice 2D, 3D T1- T2- PD- weighted 1.5T and 3T head scans acquired on Siemens and GE scanners operating with various software and hardware combinations. Results are compared against qualitative grades assigned by the Alzheimer's Disease Neuroimaging Initiative (ADNI) quality control center (taken as the reference standard). The derived quality indices are independent of the MR system used and agree with the reference standard with sensitivity and specificity higher than 90 %. With acceptable computational performance (less than 10 seconds), we incorporated our algorithm in the Siemens scanner software for routine clinical use. As for many other domains, quality criterion does not exhibit all-or-none properties (i.e., a unique quality cutoff value cannot discriminate high- versus low-quality scans) but is modulated by two main aspects addressed in this thesis: acquisition parameters (what kind of image?) and application (what for?). On the one hand, depending on acquisition parameters, more or less strong signal in the object is generated inducing more or less visible artifacts. Therefore, our quality criterion has to be adjusted accordingly. On the other hand, it appears quite natural that different quality criteria are required for algorithms segmenting global gray matter contrary to small brain structure such as hippocampus. Hence comes the second challenge addressed in this thesis: developing models that allow customizing quality criterion according to the acquisition parameters and the required performance of a target application. Taking up the first part of such challenge, we investigated certain acquisition parameters effects (i.e., echo time, bandwidth, signal suppression, field strength) and proposed an exploratory model with reasonable predictive performance (86.6 % sensitivity 92 % specificity). In order to customize quality cutoff levels according to the required performance of a target application, we developed a framework based on motion simulations. The idea is to progressively and realistically corrupt an image with motion and analyze the influence of such degradations (i.e., measured by means of our quality assessment technique) on the accuracy of a brain morphometry algorithm. To be faithful to the challenging research context of early diagnosis of AD, we targeted the automated morphometry of gray matter (GM) tissue (i.e., important imaging-based marker of pathological atrophy). Our framework reveals that 2 % error on the global volume of GM can be induced by a 1 mm patient motion halfway through the acquisition (critical low-frequency sampling of rectilinear k-space). However, this small movement is reliably caught by our quality index whose variation can be further characterized as an exponential function of the error. This model thus allows determining whether an image is of sufficient quality to warrant further quantitative analysis. In conclusion, this thesis underscores the importance of a rigorous quality assessment of MR images to support high-quality outputs of quantitative morphometry tools, ultimately increasing probability for higher diagnostic accuracy. Overall, we envision that such quality assessment could greatly improve clinical workflow through its ability to rule-out the need for a repeat scan while the patient is still in the magnet bore. Just as importantly, it would be an important step forward for incorporating morphometry analysis into regular brain-imaging protocols and thus, further exploit the clinical potential of MRI.

EPFL2011

Chargement

Afficher plus

CVSnet: A machine learning approach for automated central vein sign assessment in multiple sclerosis

Meritxell Bach Cuadra, Mario Joao Fartaria de Oliveira, João Pedro Forjaco Jorge, Cristina Granziera, Tobias Kober, Francesco La Rosa, Reto Meuli, Jonas Richiardi

2020

, ,

2010

Toward Reliable MR-Based Brain Structure Morphometry

Bénédicte Marie Maréchal

EPFL2011