Segmentation and Quantification of White Matter Lesions

White Matter Lesions

White matter lesions (WMLs) are areas of demyelinated cells found in the white matter of the brain. Minor cases that are commonly found in people over 65 years old, are thought to be the result of normal aging. However, there are other factors that contribute to the presence and amount of WMLs such as migraine headaches, stroke, or progressive neurological diseases that cause brain degeneration; such as multiple sclerosis and Alzheimer’s.

While it is not clear how or if white matter lesions directly cause brain dysfunction, they can be used as biomarkers for underlying pathology. There is a proven connection between WMLs and a decrease in brain volume, loss of memory and vision, and a decrease in cognitive ability. Studies have found that Alzheimer’s disease is likely to progresses more rapidly in patients with a greater volume of white matter lesions.

RUN DMC study

Background

Cerebral small vessel disease (SVD) is a frequent finding on CT and MRI scans of elderly people and is related to vascular risk factors and cognitive and motor impairment, ultimately leading to dementia or parkinsonism in some. In general, the relations are weak, and not all subjects with SVD become demented or get parkinsonism. This might be explained by the diversity of underlying pathology of both WML and the normal appearing white matter (NAWM). Both cannot be properly appreciated with conventional MRI. Diffusion tensor imaging (DTI) provides alternative information on microstructural white matter integrity. The association between SVD, its microstructural integrity, and incident dementia and parkinsonism has never been investigated.

Methods/Design

The RUN DMC study is a prospective cohort study on the risk factors and cognitive and motor consequences of brain changes among 503 non-demented elderly, aged between 50-85 years, with cerebral SVD. The first follow up started on July 2011. Participants alive are included and invited to the research centre to undergo a structured questionnaire on demographics and vascular risk factors, and a cognitive, and motor, assessment, followed by a MRI protocol including conventional MRI, DTI and resting state fMRI.

For all patients the WMLs are manually annotated on the FLAIR volumes. A process which is time consuming and prone to inter and intra observer variability.

alt T1andFLAIR
Figure 1: T1 (left) and FLAIR (right) MRI volumes. WML are defined as hyperintense lesions on FLAIR MRI without corresponding cerebrospinal fluid like hypo-intense lesions on the T1 weighted image. Gliosis surrounding lacunar and territorial infarcts are not considered to be WML. (See below for annotation)

alt WMLAnnotations
Figure 2: Manual WML annotations by a reader blinded to all clinical parameters

The aim of this project is to develop a solid algorithmic framework based on the analysis of different types of medical image data for improving the diagnosis of neuro-degenerative diseases, in particular dementia, Parkinsonism and decline from baseline in cognitive and motor performance. We will develop algorithms for analysing image data, extracting features from them, assessing the relevance of these features and the interplay among different types of data generated using different types of imaging technologies as well as clinical cognitive and motor performance assessment data. A prototype workstation incorporating the developed algorithms will be developed in close collaboration with the neurology and neuroradiology department at the Radboud University Nijmegen Medical Centre.

Collaborations

This is a joint research project of the Machine Learning (ML) group at iCIS, the Diagnostic Image Analysis Group (DIAG) at the Department of Radiology, and the Department of Neurology. We also have strong ties and collaborations with Fraunhofer MEVIS (Germany).

Researchers

Key publications

  • M. Ghafoorian, N. Karssemeijer, I.W.M. van Uden, F.-E. de Leeuw, T. Heskes, E. Marchiori and B. Platel. "Automated Detection of White Matter Hyperintensities of All Sizes in Cerebral Small Vessel Disease", Medical Physics 2016;43(12):6246-6258. Abstract/PDF DOI PMID

  • M. Ghafoorian, N. Karssemeijer, T. Heskes, I.W.M. van Uden, F.-E. de Leeuw, E. Marchiori, B. van Ginneken and B. Platel. "Non-uniform patch sampling with deep convolutional neural networks for white matter hyperintensity segmentation", in: IEEE International Symposium on Biomedical Imaging, 2016, pages 1414-1417. Abstract/PDF DOI

  • K. Vijverberg, M. Ghafoorian, I.W.M. van Uden, F.-E. de Leeuw, B. Platel and T. Heskes. "A single-layer network unsupervised feature learning method for white matter hyperintensity segmentation", in: Medical Imaging, 2016. Abstract/PDF DOI

  • M. Ghafoorian, N. Karssemeijer, F.E. de Leeuw, T. Heskes, E. Marchiori and B. Platel. "Small White Matter Lesion Detection in Cerebral Small Vessel Disease", in: Medical Imaging, volume 9414 of Proceedings of the SPIE, 2015, page 941411. Abstract/PDF DOI

  • M.M. Riad, B. Platel, F.-E. de Leeuw and N. Karssemeijer. "Detection of white matter lesions in cerebral small vessel disease", in: Medical Imaging, volume 8670 of Proceedings of the SPIE, 2013. Abstract DOI