Registration strategies for multi‐modal whole‐body MRI mosaicing

    loading  Checking for direct PDF access through Ovid

Excerpt

The value of anatomical whole‐body MRI is currently investigated in several pathologies, including soft tissue tumour detection 1, diagnosis, and prognosis of multiple mylenoma 2, evaluation of the lymph nodes 4 and bone marrow in pediatric age 5. In combination with functional whole‐body MR imaging, such as diffusion‐weighted MRI (DW‐MRI), it allows for lesion characterisation 6, diagnosis of bone involvement by metastases 7, and treatment response assessment 9.
Because of the MR coil size limitations, whole‐body anatomical MRI and whole‐body DWI are acquired in separate image segments. Depending on the acquisition, these consist of slabs of two‐dimensional (2D) axial slices or three‐dimensional (3D) volumes (anatomical sequences) called image stations, which are later combined into a whole‐body image. Image stations suffer from inter‐station intensity variations 11, which have to be corrected in order to obtain a homogeneous whole‐body image.
Such intensity standardization between stations has been described previously and will not be considered in this work 13. However, the whole‐body MR images often bear additional artifacts at the transition between adjacent stations, caused by patient motion between the acquisition of subsequent stations, and leading to inconsistent anatomical views between stations. Other sources of artifacts include geometric distortions and image shearing caused by eddy currents, induced by rapid switching of gradient coils in echo‐planar MR imaging (EPI) sequence; and voxel shifts in the phase‐encoding direction in DW images, caused by differences in applied frequency offsets 14. Compensation of such artifacts is the subject of this study.
Geometric artifacts may result in inaccurate representations of the anatomy and yield unreliable morphological measurements, including the size and volume of organs or tumours, or affect the assessment of functional quantities, such as the global apparent diffusion coefficient (ADC) 9. Moreover, the lack of registration between anatomical and functional images may limit or lead to false results in DWI parameters measurements performed on the basis of region of interest defined on anatomical images. It may also compromise the quality of inter‐station intensity standardization (ISIS) algorithms—usually driven by the registration of intensity histograms acquired from the image stations or the overlap between the neighbouring image stations 13. In addition, misalignment between multi‐modal whole‐body MR data hinders visual fusion of the image modalities, thereby limiting the benefits of the combined inspection of function and anatomy.
To compensate for the aforementioned inconsistencies, image stations can be acquired including an overlap of a predefined length between neighbouring segments. Mosaicing of the whole‐body image stations and aligning them to their multi‐modal corresponding image using image registration remains a challenging task, in particular for DW‐MRI. These functional images have relatively low spatial resolution (slice thicknesses above 5 mm are common), poor signal‐to‐noise ratio (SNR) and the overlapping regions are often limited to a few slices 15. In addition, prior to ISIS, voxel intensities cannot be compared directly for whole‐body mosaicing or alignment with other MR modalities. In many cases, whole‐body anatomical (e.g., T1 or T2‐weighted) MRI is acquired during the same session. Such anatomical MR images have comparatively high spatial resolution and are not affected by DWI related image distortions, which make them reliable anatomical reference images.
To date, three approaches have been proposed aimed at solving related registration problems to the one described above. Blackledge et al. 9 proposed a sequential station‐to‐station registration algorithm for diffusion‐weighted MR, compensating for uni‐dimensional translations along phase‐encoding direction. The registration was applied to DW b1000 image stations, and driven by a mean squared differences (MSD) metric after applying ISIS. Alignment with the anatomical MRI was not addressed.

Related Topics

    loading  Loading Related Articles