Automatic histogram-based segmentation of white matter hyperintensities using 3D FLAIR images

Rita Lopes Simoes, Cornelis H. Slump, Christoph Mönninghoff, Isabel Wanke, Martha Dlugaj, Christian Weimar

    Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademic

    1 Citation (Scopus)
    1 Downloads (Pure)


    White matter hyperintensities are known to play a role in the cognitive decline experienced by patients suffering from neurological diseases. Therefore, accurately detecting and monitoring these lesions is of importance. Automatic methods for segmenting white matter lesions typically use multimodal MRI data. Furthermore, many methods use a training set to perform a classi﬿cation task or to determine necessary parameters. In this work, we describe and evaluate an unsupervised segmentation method that is based solely on the histogram of FLAIR images. It approximates the histogram by a mixture of three Gaussians in order to ﬿nd an appropriate threshold for white matter hyperintensities. We use a context-sensitive Expectation-Maximization method to determine the Gaussian mixture parameters. The segmentation is subsequently corrected for false positives using the knowledge of the location of typical FLAIR artifacts. A preliminary validation with the ground truth on 6 patients revealed a Similarity Index of 0.73 ± 0.10, indicating that the method is comparable to others in the literature which require multimodal MRI and/or a preliminary training step.
    Original languageUndefined
    Title of host publicationMedical Imaging 2012: Computer-Aided Diagnosis
    EditorsBram van Ginneken, Carol L. Novak
    Number of pages10
    ISBN (Print)978-0-81948-964-7
    Publication statusPublished - 2012
    EventSPIE Medical Imaging 2012 - Town & Country Resort and Convention Center, San Diego, United States
    Duration: 4 Feb 20129 Feb 2012

    Publication series

    NameProceedings of SPIE


    ConferenceSPIE Medical Imaging 2012
    Country/TerritoryUnited States
    CitySan Diego


    • IR-79875
    • EWI-21632
    • METIS-285168

    Cite this