Automatic multiclass intramedullary spinal cord tumor segmentation on MRI with deep learning.

Spinal cord tumors lead to neurological morbidity and mortality. Being able to obtain morphometric quantification (size, location, growth rate) of the tumor, edema, and cavity can result in improved monitoring and treatment planning. Such quantification requires the segmentation of these structures into three separate classes. However, manual segmentation of three-dimensional structures is time consuming, tedious and prone to intra- and inter-rater variability, motivating the development of automated methods. Here, we tailor a model adapted to the spinal cord tumor segmentation task. Data were obtained from 343 patients using gadolinium-enhanced T1-weighted and T2-weighted MRI scans with cervical, thoracic, and/or lumbar coverage. The dataset includes the three most common intramedullary spinal cord tumor types: astrocytomas, ependymomas, and hemangioblastomas. The proposed approach is a cascaded architecture with U-Net-based models that segments tumors in a two-stage process: locate and label. The model first finds the spinal cord and generates bounding box coordinates. The images are cropped according to this output, leading to a reduced field of view, which mitigates class imbalance. The tumor is then segmented. The segmentation of the tumor, cavity, and edema (as a single class) reached 76.7 ± 1.5% of Dice score and the segmentation of tumors alone reached 61.8 ± 4.0% Dice score. The true positive detection rate was above 87% for tumor, edema, and cavity. To the best of our knowledge, this is the first fully automatic deep learning model for spinal cord tumor segmentation. The multiclass segmentation pipeline is available in the Spinal Cord Toolbox (https://spinalcordtoolbox.com/). It can be run with custom data on a regular computer within seconds.

Open-access quantitative MRI data of the spinal cord and reproducibility across participants, sites and manufacturers.

In a companion paper by Cohen-Adad et al. we introduce the spine generic quantitative MRI protocol that provides valuable metrics for assessing spinal cord macrostructural and microstructural integrity. This protocol was used to acquire a single subject dataset across 19 centers and a multi-subject dataset across 42 centers (for a total of 260 participants), spanning the three main MRI manufacturers: GE, Philips and Siemens. Both datasets are publicly available via git-annex. Data were analysed using the Spinal Cord Toolbox to produce normative values as well as inter/intra-site and inter/intra-manufacturer statistics. Reproducibility for the spine generic protocol was high across sites and manufacturers, with an average inter-site coefficient of variation of less than 5% for all the metrics. Full documentation and results can be found at https://spine-generic.rtfd.io/ . The datasets and analysis pipeline will help pave the way towards accessible and reproducible quantitative MRI in the spinal cord.

Generic acquisition protocol for quantitative MRI of the spinal cord.

Quantitative spinal cord (SC) magnetic resonance imaging (MRI) presents many challenges, including a lack of standardized imaging protocols. Here we present a prospectively harmonized quantitative MRI protocol, which we refer to as the spine generic protocol, for users of 3T MRI systems from the three main manufacturers: GE, Philips and Siemens. The protocol provides guidance for assessing SC macrostructural and microstructural integrity: T1-weighted and T2-weighted imaging for SC cross-sectional area computation, multi-echo gradient echo for gray matter cross-sectional area, and magnetization transfer and diffusion weighted imaging for assessing white matter microstructure. In a companion paper from the same authors, the spine generic protocol was used to acquire data across 42 centers in 260 healthy subjects. The key details of the spine generic protocol are also available in an open-access document that can be found at https://github.com/spine-generic/protocols . The protocol will serve as a starting point for researchers and clinicians implementing new SC imaging initiatives so that, in the future, inclusion of the SC in neuroimaging protocols will be more common. The protocol could be implemented by any trained MR technician or by a researcher/clinician familiar with MRI acquisition.

SoftSeg: Advantages of soft versus binary training for image segmentation.

Most image segmentation algorithms are trained on binary masks formulated as a classification task per pixel. However, in applications such as medical imaging, this "black-and-white" approach is too constraining because the contrast between two tissues is often ill-defined, i.e., the voxels located on objects' edges contain a mixture of tissues (a partial volume effect). Consequently, assigning a single "hard" label can result in a detrimental approximation. Instead, a soft prediction containing non-binary values would overcome that limitation. In this study, we introduce SoftSeg, a deep learning training approach that takes advantage of soft ground truth labels, and is not bound to binary predictions. SoftSeg aims at solving a regression instead of a classification problem. This is achieved by using (i) no binarization after preprocessing and data augmentation, (ii) a normalized ReLU final activation layer (instead of sigmoid), and (iii) a regression loss function (instead of the traditional Dice loss). We assess the impact of these three features on three open-source MRI segmentation datasets from the spinal cord gray matter, the multiple sclerosis brain lesion, and the multimodal brain tumor segmentation challenges. Across multiple random dataset splittings, SoftSeg outperformed the conventional approach, leading to an increase in Dice score of 2.0% on the gray matter dataset (p=0.001), 3.3% for the brain lesions, and 6.5% for the brain tumors. SoftSeg produces consistent soft predictions at tissues' interfaces and shows an increased sensitivity for small objects (e.g., multiple sclerosis lesions). The richness of soft labels could represent the inter-expert variability, the partial volume effect, and complement the model uncertainty estimation, which is typically unclear with binary predictions. The developed training pipeline can easily be incorporated into most of the existing deep learning architectures. SoftSeg is implemented in the freely-available deep learning toolbox ivadomed (https://ivadomed.org).

A Cross-Sectional Study on the Impact of Arterial Stiffness on the Corpus Callosum, a Key White Matter Tract Implicated in Alzheimer's Disease.

BACKGROUND: Vascular risk factors such as arterial stiffness play an important role in the etiology of Alzheimer's disease (AD), presumably due to the emergence of white matter lesions. However, the impact of arterial stiffness to white matter structure involved in the etiology of AD, including the corpus callosum remains poorly understood. OBJECTIVE: The aims of the study are to better understand the relationship between arterial stiffness, white matter microstructure, and perfusion of the corpus callosum in older adults. METHODS: Arterial stiffness was estimated using the gold standard measure of carotid-femoral pulse wave velocity (cfPWV). Cognitive performance was evaluated with the Trail Making Test part B-A. Neurite orientation dispersion and density imaging was used to obtain microstructural information such as neurite density and extracellular water diffusion. The cerebral blood flow was estimated using arterial spin labelling. RESULTS: cfPWV better predicts the microstructural integrity of the corpus callosum when compared with other index of vascular aging (the augmentation index, the systolic blood pressure, and the pulse pressure). In particular, significant associations were found between the cfPWV, an alteration of the extracellular water diffusion, and a neuronal density increase in the body of the corpus callosum which was also correlated with the performance in cognitive flexibility. CONCLUSION: Our results suggest that arterial stiffness is associated with an alteration of brain integrity which impacts cognitive function in older adults.

Multiple sclerosis lesions in motor tracts from brain to cervical cord: spatial distribution and correlation with disability.

Despite important efforts to solve the clinico-radiological paradox, correlation between lesion load and physical disability in patients with multiple sclerosis remains modest. One hypothesis could be that lesion location in corticospinal tracts plays a key role in explaining motor impairment. In this study, we describe the distribution of lesions along the corticospinal tracts from the cortex to the cervical spinal cord in patients with various disease phenotypes and disability status. We also assess the link between lesion load and location within corticospinal tracts, and disability at baseline and 2-year follow-up. We retrospectively included 290 patients (22 clinically isolated syndrome, 198 relapsing remitting, 39 secondary progressive, 31 primary progressive multiple sclerosis) from eight sites. Lesions were segmented on both brain (T2-FLAIR or T2-weighted) and cervical (axial T2- or T2*-weighted) MRI scans. Data were processed using an automated and publicly available pipeline. Brain, brainstem and spinal cord portions of the corticospinal tracts were identified using probabilistic atlases to measure the lesion volume fraction. Lesion frequency maps were produced for each phenotype and disability scores assessed with Expanded Disability Status Scale score and pyramidal functional system score. Results show that lesions were not homogeneously distributed along the corticospinal tracts, with the highest lesion frequency in the corona radiata and between C2 and C4 vertebral levels. The lesion volume fraction in the corticospinal tracts was higher in secondary and primary progressive patients (mean = 3.6 ± 2.7% and 2.9 ± 2.4%), compared to relapsing-remitting patients (1.6 ± 2.1%, both P < 0.0001). Voxel-wise analyses confirmed that lesion frequency was higher in progressive compared to relapsing-remitting patients, with significant bilateral clusters in the spinal cord corticospinal tracts (P < 0.01). The baseline Expanded Disability Status Scale score was associated with lesion volume fraction within the brain (r = 0.31, P < 0.0001), brainstem (r = 0.45, P < 0.0001) and spinal cord (r = 0.57, P < 0.0001) corticospinal tracts. The spinal cord corticospinal tracts lesion volume fraction remained the strongest factor in the multiple linear regression model, independently from cord atrophy. Baseline spinal cord corticospinal tracts lesion volume fraction was also associated with disability progression at 2-year follow-up (P = 0.003). Our results suggest a cumulative effect of lesions within the corticospinal tracts along the brain, brainstem and spinal cord portions to explain physical disability in multiple sclerosis patients, with a predominant impact of intramedullary lesions.

Open-source pipeline for multi-class segmentation of the spinal cord with deep learning.

This paper presents an open-source pipeline to train neural networks to segment structures of interest from MRI data. The pipeline is tailored towards homogeneous datasets and requires relatively low amounts of manual segmentations (few dozen, or less depending on the homogeneity of the dataset). Two use-case scenarios for segmenting the spinal cord white and grey matter are presented: one in marmosets with variable numbers of lesions, and the other in the publicly available human grey matter segmentation challenge [1]. The pipeline is freely available at: https://github.com/neuropoly/multiclass-segmentation.

Spatial distribution of multiple sclerosis lesions in the cervical spinal cord.

Spinal cord lesions detected on MRI hold important diagnostic and prognostic value for multiple sclerosis. Previous attempts to correlate lesion burden with clinical status have had limited success, however, suggesting that lesion location may be a contributor. Our aim was to explore the spatial distribution of multiple sclerosis lesions in the cervical spinal cord, with respect to clinical status. We included 642 suspected or confirmed multiple sclerosis patients (31 clinically isolated syndrome, and 416 relapsing-remitting, 84 secondary progressive, and 73 primary progressive multiple sclerosis) from 13 clinical sites. Cervical spine lesions were manually delineated on T2- and T2*-weighted axial and sagittal MRI scans acquired at 3 or 7 T. With an automatic publicly-available analysis pipeline we produced voxelwise lesion frequency maps to identify predilection sites in various patient groups characterized by clinical subtype, Expanded Disability Status Scale score and disease duration. We also measured absolute and normalized lesion volumes in several regions of interest using an atlas-based approach, and evaluated differences within and between groups. The lateral funiculi were more frequently affected by lesions in progressive subtypes than in relapsing in voxelwise analysis (P < 0.001), which was further confirmed by absolute and normalized lesion volumes (P < 0.01). The central cord area was more often affected by lesions in primary progressive than relapse-remitting patients (P < 0.001). Between white and grey matter, the absolute lesion volume in the white matter was greater than in the grey matter in all phenotypes (P < 0.001); however when normalizing by each region, normalized lesion volumes were comparable between white and grey matter in primary progressive patients. Lesions appearing in the lateral funiculi and central cord area were significantly correlated with Expanded Disability Status Scale score (P < 0.001). High lesion frequencies were observed in patients with a more aggressive disease course, rather than long disease duration. Lesions located in the lateral funiculi and central cord area of the cervical spine may influence clinical status in multiple sclerosis. This work shows the added value of cervical spine lesions, and provides an avenue for evaluating the distribution of spinal cord lesions in various patient groups.

Automatic segmentation of the spinal cord and intramedullary multiple sclerosis lesions with convolutional neural networks.

The spinal cord is frequently affected by atrophy and/or lesions in multiple sclerosis (MS) patients. Segmentation of the spinal cord and lesions from MRI data provides measures of damage, which are key criteria for the diagnosis, prognosis, and longitudinal monitoring in MS. Automating this operation eliminates inter-rater variability and increases the efficiency of large-throughput analysis pipelines. Robust and reliable segmentation across multi-site spinal cord data is challenging because of the large variability related to acquisition parameters and image artifacts. In particular, a precise delineation of lesions is hindered by a broad heterogeneity of lesion contrast, size, location, and shape. The goal of this study was to develop a fully-automatic framework - robust to variability in both image parameters and clinical condition - for segmentation of the spinal cord and intramedullary MS lesions from conventional MRI data of MS and non-MS cases. Scans of 1042 subjects (459 healthy controls, 471 MS patients, and 112 with other spinal pathologies) were included in this multi-site study (n = 30). Data spanned three contrasts (T1-, T2-, and T2∗-weighted) for a total of 1943 vol and featured large heterogeneity in terms of resolution, orientation, coverage, and clinical conditions. The proposed cord and lesion automatic segmentation approach is based on a sequence of two Convolutional Neural Networks (CNNs). To deal with the very small proportion of spinal cord and/or lesion voxels compared to the rest of the volume, a first CNN with 2D dilated convolutions detects the spinal cord centerline, followed by a second CNN with 3D convolutions that segments the spinal cord and/or lesions. CNNs were trained independently with the Dice loss. When compared against manual segmentation, our CNN-based approach showed a median Dice of 95% vs. 88% for PropSeg (p ≤ 0.05), a state-of-the-art spinal cord segmentation method. Regarding lesion segmentation on MS data, our framework provided a Dice of 60%, a relative volume difference of -15%, and a lesion-wise detection sensitivity and precision of 83% and 77%, respectively. In this study, we introduce a robust method to segment the spinal cord and intramedullary MS lesions on a variety of MRI contrasts. The proposed framework is open-source and readily available in the Spinal Cord Toolbox.

Automatic spinal cord localization, robust to MRI contrasts using global curve optimization.

During the last two decades, MRI has been increasingly used for providing valuable quantitative information about spinal cord morphometry, such as quantification of the spinal cord atrophy in various diseases. However, despite the significant improvement of MR sequences adapted to the spinal cord, automatic image processing tools for spinal cord MRI data are not yet as developed as for the brain. There is nonetheless great interest in fully automatic and fast processing methods to be able to propose quantitative analysis pipelines on large datasets without user bias. The first step of most of these analysis pipelines is to detect the spinal cord, which is challenging to achieve automatically across the broad range of MRI contrasts, field of view, resolutions and pathologies. In this paper, a fully automated, robust and fast method for detecting the spinal cord centerline on MRI volumes is introduced. The algorithm uses a global optimization scheme that attempts to strike a balance between a probabilistic localization map of the spinal cord center point and the overall spatial consistency of the spinal cord centerline (i.e. the rostro-caudal continuity of the spinal cord). Additionally, a new post-processing feature, which aims to automatically split brain and spine regions is introduced, to be able to detect a consistent spinal cord centerline, independently from the field of view. We present data on the validation of the proposed algorithm, known as "OptiC", from a large dataset involving 20 centers, 4 contrasts (T2-weighted n = 287, T1-weighted n = 120, T2∗-weighted n = 307, diffusion-weighted n = 90), 501 subjects including 173 patients with a variety of neurologic diseases. Validation involved the gold-standard centerline coverage, the mean square error between the true and predicted centerlines and the ability to accurately separate brain and spine regions. Overall, OptiC was able to cover 98.77% of the gold-standard centerline, with a mean square error of 1.02 mm. OptiC achieved superior results compared to a state-of-the-art spinal cord localization technique based on the Hough transform, especially on pathological cases with an averaged mean square error of 1.08 mm vs. 13.16 mm (Wilcoxon signed-rank test p-value < .01). Images containing brain regions were identified with a 99% precision, on which brain and spine regions were separated with a distance error of 9.37 mm compared to ground-truth. Validation results on a challenging dataset suggest that OptiC could reliably be used for subsequent quantitative analyses tasks, opening the door to more robust analysis on pathological cases.