PET/MRI attenuation correction

PET is an important non-invasive imaging tool in modern medicine to aid clinical decision-making. The technique visualizes functional biological processes, by detecting radioactive decay of an injected radiotracer as it accumulates within the patient. The decay results in the emission of two gamma photons that can be detected by the scanner as they escape the patient. The likelihood of them escaping depends on the material they must pass through, as dense material (e.g., bone tissue) attenuates the photons more than non-dense material (e.g., soft tissue). Therefore, to achieve accurate PET images, it is important to correct for photon attenuation, for which a mapping of the attenuating properties of the patient's body – an attenuation map – is required. In a dual-modality PET/CT scanner, the attenuation map can easily be obtained from the CT images; however, such information is not available in PET/MRI, since the MR image cannot easily be converted to an attenuation map. Instead, attenuation maps has to be derived or synthesized from the available PET or MR images – a task that has proven to be one of the most challenging issues with PET/MRI. The methods that have been proposed to remedy the problem of attenuation correction in the last decade can largely be grouped into four major categories: (1) segmentation; division of MR images into pre-defined classes each representing an underlying tissue type, (2) atlas; computation of a synthetic CT by aligning a database of MRI–CT pairs, (3) emission; simultaneous reconstruction of attenuation map and attenuation corrected PET image directly from the PET data, and (4) machine learning; training a model to learn a mapping from MRI to CT. This chapter discusses the challenges associated with PET/MRI attenuation correction and gives an overview of the proposed solutions developed within each of the four categories, from the earliest methods that surfaced at the introduction of the first clinical system to state-of-the-art methods.