NEArBy : lexical normalization for atlas enabled CBR system for neuroimaging

Abstract

Brain atlases have been used as spatial references to classify and tag either structural or functional topological information from brain images. Semantic information obtained from the existing image data is thus spatially mapped according the atlas descriptors. However the process of classifying and tagging brain images using an atlas is often tedious and mostly dependent on human observation and validation. At the same time, even when available, it is often difficult to use, particularly when using standard query and retrieve services in modern imaging repositories (e.g. DICOM based PACS). In this work we propose NEArBy, a cloud based solution that provides query and retrieve services based on brain atlas semantics that can be easily integrated in existing DICOM based imaging repositories. Using a web interface, NEArBy supports not only typical DICOM query retrieve searches but also query tokens matching the brain atlas dictionary. To automate the semantic tagging of the brain images we rely on external methods to identify relevant spatial features that are later labelled using standard brain atlas. Being DICOM a tag based standard, atlas related tags are then privately embedded into DICOM files as NEArBy JSON descriptors using lexicon as proposed in NeuroLex. These descriptors encode the mapping between feature type, spatial location in the atlas and the respective atlas tag. JSON encoded tags are also suitable for indexing by a medical imaging Q/R tool such as Dicoogle allowing queries based both on standard DICOM tags and specifically on atlas related tokens included by NEArBy middleware. NEArBy provides a new way to perform non- patient centric queries over neuro-imaging repositories using technical and atlas based topological information. During this dissertation, the NEArBy potential usage is illustrated over a set of functional magnetic resonance imaging (fMRI) datasets using the web user interface to formulate the queries with atlas related criteria and access the retrieved results.

Brain atlases have been used as spatial references to classify and tag either structural or functional topological information from brain images. Semantic information obtained from the existing image data is thus spatially mapped according the atlas descriptors. However the process of classifying and tagging brain images using an atlas is often tedious and mostly dependent on human observation and validation. At the same time, even when available, it is often difficult to use, particularly when using standard query and retrieve services in modern imaging repositories (e.g. DICOM based PACS). In this work we propose NEArBy, a cloud based solution that provides query and retrieve services based on brain atlas semantics that can be easily integrated in existing DICOM based imaging repositories. Using a web interface, NEArBy supports not only typical DICOM query retrieve searches but also query tokens matching the brain atlas dictionary. To automate the semantic tagging of the brain images we rely on external methods to identify relevant spatial features that are later labelled using standard brain atlas. Being DICOM a tag based standard, atlas related tags are then privately embedded into DICOM files as NEArBy JSON descriptors using lexicon as proposed in NeuroLex. These descriptors encode the mapping between feature type, spatial location in the atlas and the respective atlas tag. JSON encoded tags are also suitable for indexing by a medical imaging Q/R tool such as Dicoogle allowing queries based both on standard DICOM tags and specifically on atlas related tokens included by NEArBy middleware. NEArBy provides a new way to perform non- patient centric queries over neuro-imaging repositories using technical and atlas based topological information. During this dissertation, the NEArBy potential usage is illustrated over a set of functional magnetic resonance imaging (fMRI) datasets using the web user interface to formulate the queries with atlas related criteria and access the retrieved results.