Multicenter Evaluation of Geometric Accuracy of MRI Protocols Used in Experimental Stroke

Abstract

It has recently been suggested that multicenter preclinical stroke studies should be carried out to improve translation from bench to bedside, but the accuracy of magnetic resonance imaging (MRI) scanners routinely used in experimental stroke has not yet been evaluated. We aimed to assess and compare geometric accuracy of preclinical scanners and examine the longitudinal stability of one scanner using a simple quality assurance (QA) protocol. Six 7 Tesla animal scanners across six different preclinical imaging centers throughout Europe were used to scan a small structural phantom and estimate linear scaling errors in all orthogonal directions and volumetric errors. Between-scanner imaging consisted of a standard sequence and each center’s preferred sequence for the assessment of infarct size in rat models of stroke. The standard sequence was also used to evaluate the drift in accuracy of the worst performing scanner over a period of six months following basic gradient calibration. Scaling and volumetric errors using the standard sequence were less variable than corresponding errors using different stroke sequences. The errors for one scanner, estimated using the standard sequence, were very high (above 4% scaling errors for each orthogonal direction, 18.73% volumetric error). Calibration of the gradient coils in this system reduced scaling errors to within ±1.0%; these remained stable during the subsequent 6-month assessment. In conclusion, despite decades of use in experimental studies, preclinical MRI still suffers from poor and variable geometric accuracy, influenced by the use of miscalibrated systems and various types of sequences for the same purpose. For effective pooling of data in multicenter studies, centers should adopt standardized procedures for system QA and in vivo imaging.

In vivo imaging biomarkers of neuroinflammation in the development and assessment of stroke therapies - towards clinical translation

“…Despite the multitude of advances in imaging-guided therapies, there is still a strong need to further improve the quality of research to obtain predictive and clinically translatable data. To this end, research activities should involve collaborative community efforts, including multi-center studies as suggested 10 , 194 and applied recently 54 , 195 - 198 . The concept of reverse-translating quality standards from clinical studies into preclinical work has shown promise and overall acceptance to further enhance translatability of research 199 .…”

Atlas registration for edema-corrected MRI lesion volume in mouse stroke models

“…Since we only recently became aware of the problem of non-whole brain coverage for MLC, sham animals were scanned using our old standard protocol with 20 slices covering mid- and forebrain only. 27 In plane directions are termed x (lateral) and y (ventral-dorsal), whereas slice selection direction z is referring to the caudal-rostral axis.…”

Adaptive scale based u-shape transformer network for ischemic stroke lesion segmentation in CTP images

“…When computing the loss, we interpret the samples in the batch as a whole sample and calculate the dice loss over all voxels in the cluster. The dice loss formulation used here is proposed by Fabian et al, which is adapted from the multi-class adaptation of the variant proposed by M.Drozdzal et al [23] The implementation of dice loss is as follows: [23] The ISLES 2018 Challenge tackled the segmentation of stroke lesion core from acute CT scan. Imaging data from acute stroke patients in two centers is presented within 8 hours of stroke onset.…”