Location Sensitive Deep Convolutional Neural Networks for Segmentation of White Matter Hyperintensities - PubMed Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Jul 11;7(1):5110.
doi: 10.1038/s41598-017-05300-5.

Location Sensitive Deep Convolutional Neural Networks for Segmentation of White Matter Hyperintensities

Mohsen Ghafoorian  1   2 , Nico Karssemeijer  3 , Tom Heskes  4 , Inge W M van Uden  5 , Clara I Sanchez  3 , Geert Litjens  3 , Frank-Erik de Leeuw  5 , Bram van Ginneken  3 , Elena Marchiori  4 , Bram Platel  3
Affiliations

Location Sensitive Deep Convolutional Neural Networks for Segmentation of White Matter Hyperintensities

Mohsen Ghafoorian et al. Sci Rep. .

Abstract

The anatomical location of imaging features is of crucial importance for accurate diagnosis in many medical tasks. Convolutional neural networks (CNN) have had huge successes in computer vision, but they lack the natural ability to incorporate the anatomical location in their decision making process, hindering success in some medical image analysis tasks. In this paper, to integrate the anatomical location information into the network, we propose several deep CNN architectures that consider multi-scale patches or take explicit location features while training. We apply and compare the proposed architectures for segmentation of white matter hyperintensities in brain MR images on a large dataset. As a result, we observe that the CNNs that incorporate location information substantially outperform a conventional segmentation method with handcrafted features as well as CNNs that do not integrate location information. On a test set of 50 scans, the best configuration of our networks obtained a Dice score of 0.792, compared to 0.805 for an independent human observer. Performance levels of the machine and the independent human observer were not statistically significantly different (p-value = 0.06).

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Figure 1
Figure 1
A pattern is observable in WMHs occurrence probability map.
Figure 2
Figure 2
An example of negative (top row) and positive (bottom row) samples in three scales (from left to right) 32 × 32, 64 × 64 and 128 × 128 on the FLAIR image. The two larger scales are down sampled to 32 × 32.
Figure 3
Figure 3
Patch preparation process and different proposed CNN architectures. The links between set of convolutional layers represent a weight sharing policy among the streams.
Figure 4
Figure 4
Integration of spatial location information fills the gap between performance of a normal CNN and human observer. (a) An ROC comparison of different CNN methods, a conventional segmentation method and independent human observer, considering observer 1 as the reference standard. (b) A comparison of different methods on Dice score as a function of binary masking threshold. The light shades around the curves indicate 95% confidence intervals with bootstrapping on patients.
Figure 5
Figure 5
Test Dice as a function of training set size.
Figure 6
Figure 6
Two sample cases of segmentation improvement by adding location information to the network. (a) FLAIR images without annotations. (b) Segmentation by human observer 1. (c) Segmentation by SS method. (d) Segmentation by MSWS + Loc method.
Figure 7
Figure 7
Gliosis around the lacunes is a prevalent type of false positive segmentation. (a) FLAIR images without annotations. (b) Segmentation by human observer 1. (c) Segmentation by human observer 2. (d) Segmentation by MSWS + Loc method.
Figure 8
Figure 8
A sample case with a small lesion missed by the two human observers. (a) FLAIR image without annotations. (b) Segmentation by human observer 1. (c) Segmentation by human observer 2. (d) Segmentation by MSWS + Loc method.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. van Norden AG, et al. Causes and consequences of cerebral small vessel disease. The RUN DMC study: a prospective cohort study. Study rationale and protocol. BMC Neurol. 2011;11:29. doi: 10.1186/1471-2377-11-29. - DOI - PMC - PubMed
    1. Schoonheim MM, et al. Sex-specific extent and severity of white matter damage in multiple sclerosis: Implications for cognitive decline. Human Brain Mapping. 2014;35:2348–2358. doi: 10.1002/hbm.22332. - DOI - PMC - PubMed
    1. Marshall G, Shchelchkov E, Kaufer D, Ivanco L, Bohnen N. White matter hyperintensities and cortical acetylcholinesterase activity in parkinsonian dementia. Acta Neurologica Scandinavica. 2006;113:87–91. doi: 10.1111/j.1600-0404.2005.00553.x. - DOI - PubMed
    1. Weinstein G, et al. Brain imaging and cognitive predictors of stroke and alzheimer disease in the framingham heart study. Stroke. 2013;44:2787–2794. doi: 10.1161/STROKEAHA.113.000947. - DOI - PMC - PubMed
    1. Hirono N, Kitagaki H, Kazui H, Hashimoto M, Mori E. Impact of white matter changes on clinical manifestation of alzheimer’s disease a quantitative study. Stroke. 2000;31:2182–2188. doi: 10.1161/01.STR.31.9.2182. - DOI - PubMed

Publication types