Segmentation of Positron Emission Tomography Images Using Multi-atlas Anatomical Magnetic Resonance Imaging (MRI)

Positron emission tomography (PET), is a medical imaging technique, it provides information about the body’s cellular function rather than its anatomy. However, due to the functional nature of PET images, locating the anatomical structures in such an image remains a challenging task, indeed, PET images only provide very little anatomical information. Segmentation of PET images, therefore, requires the intervention of a medical expert. The expert proceeds to a manual segmentation of a volume slice by slice, which turns out to be very tedious and costly in terms of time. In this article, we present, evaluate, and make available a multi-atlas approach for automatically segmenting human brain PET image combining both the information provided by the PET volume to be segmented and prior knowledge of the volume provided in the form of multi-anatomical atlas. This also performs comparably to single atlas extraction, multi-atlas methods to improve the accuracy of the defined region. As shown in this study, we achieved significant improvement after the integration of this approach with two widely used multi atlas based segmentation (MAS) methods on BIC database provided by McConnel Brain Imaging Center Montréal and LONI database (USC Neurological Imaging Laboratory).