Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2018 Apr 13;8(1):5966.
doi: 10.1038/s41598-018-24304-3.

Spinal Cord Gray Matter Segmentation Using Deep Dilated Convolutions

Affiliations
Free PMC article

Spinal Cord Gray Matter Segmentation Using Deep Dilated Convolutions

Christian S Perone et al. Sci Rep. .
Free PMC article

Abstract

Gray matter (GM) tissue changes have been associated with a wide range of neurological disorders and were recently found relevant as a biomarker for disability in amyotrophic lateral sclerosis. The ability to automatically segment the GM is, therefore, an important task for modern studies of the spinal cord. In this work, we devise a modern, simple and end-to-end fully-automated human spinal cord gray matter segmentation method using Deep Learning, that works both on in vivo and ex vivo MRI acquisitions. We evaluate our method against six independently developed methods on a GM segmentation challenge. We report state-of-the-art results in 8 out of 10 evaluation metrics as well as major network parameter reduction when compared to the traditional medical imaging architectures such as U-Nets.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1
Figure 1
In vivo axial-slice samples from four centers (UCL, Montreal, Zurich, Vanderbilt) that collaborated to the SCGM Segmentation Challenge. Top row: original MRI images. Bottom row: a crop of the spinal cord (green rectangle).
Figure 2
Figure 2
Dilated convolution. On the left, we have the dilated convolution with dilation rate r = 1, equivalent to the standard convolution. In the middle with have a dilation r = 2 and in the right a dilation rate of r = 3. All dilated convolutions have a 3 × 3 kernel size and the same number of parameters.
Figure 3
Figure 3
Architecture overview of the proposed method. The MRI axial slice is fed to the first block of 3 × 3 convolutions and then to a block of dilated convolutions (rate 2). Then, six parallel modules with different rates (6/12/18/24), 1 × 1 convolution, and a global average pooling are used in parallel. After the parallel modules, all feature maps are concatenated and then fed into the final block of 1 × 1 convolutions to produce the final dense predictions.
Figure 4
Figure 4
Architecture pipeline overview. During the first stage, input axial slices are resampled to a common pixel size space, then intensity is normalized, followed by the network inference.
Figure 5
Figure 5
Qualitative evaluation of our proposed approach on the same axial slice for subject 11 of each site. From top to bottom row: input image, majority voting segmentation gold standard, and the result of our segmentation method. Adapted from the work.
Figure 6
Figure 6
Test set evaluation results from the SCGM segmentation challenge for each evaluated metric, with the Dice similarity coefficient (DSC), mean surface distance (MSD), Hausdorff surface distance (HSD), skeletonized Hausdorff distance (SHD), skeletonized median distance (SMD), true positive rate (TPR), true negative rate (TNR), positive predictive value (PPV), Jaccard index (JI) and conformity coefficient (CC). Our method is shown as “Proposed”. Best viewed in color.
Figure 7
Figure 7
Qualitative evaluation of the U-Net and our proposed method on the ex vivo high-resolution spinal cord dataset. Each column represents a random sample of the test set (regions from left to right: sacral, thoracic, cervical). Green rectangles frame the oversegmentations of the U-Net model predictions.
Figure 8
Figure 8
Lumbosacral region 3D rendered view of the ex vivo high-resolution spinal cord dataset segmented using the proposed method. The gray matter is depicted in orange color while the white matter and other tissues are represented in transparent gray color.

Similar articles

See all similar articles

Cited by 11 articles

See all "Cited by" articles

References

    1. Amukotuwa, S. A. & Cook, M. J. Spinal Disease: Neoplastic, Degenerative, and Infective Spinal Cord Diseases and Spinal Cord Compression. Clinical Gate 511–538, 10.1016/B978-0-323-03354-1.50044-4 (2007).
    1. Schlaeger R, et al. Spinal cord gray matter atrophy correlates with multiple sclerosis disability. Annals of Neurology. 2014;76:568–580. doi: 10.1002/ana.24241. - DOI - PMC - PubMed
    1. Paquin, M.-Ê. et al. Spinal Cord Gray Matter Atrophy in Amyotrophic Lateral Sclerosis. American Journal of Neuroradiology10.3174/ajnr.A5427 (2017). - PubMed
    1. De Leener B, Taso M, Cohen-Adad J, Callot V. Segmentation of the human spinal cord. Magnetic Resonance Materials in Physics, Biology and Medicine. 2016;29:125–153. doi: 10.1007/s10334-015-0507-2. - DOI - PubMed
    1. De Leener B, Kadoury S, Cohen-Adad J. Robust, accurate and fast automatic segmentation of the spinal cord. NeuroImage. 2014;98:528–536. doi: 10.1016/j.neuroimage.2014.04.051. - DOI - PubMed

Publication types

Feedback