Sensitivity of portable low-field magnetic resonance imaging for multiple sclerosis lesions

Abstract

Magnetic resonance imaging (MRI) is a fundamental tool in the diagnosis and management of neurological diseases such as multiple sclerosis (MS). New portable, low-field strength, MRI scanners could potentially lower financial and technical barriers to neuroimaging and reach underserved or disabled populations, but the sensitivity of these devices for MS lesions is unknown. We sought to determine if white matter lesions can be detected on a portable 64mT scanner, compare automated lesion segmentations and total lesion volume between paired 3T and 64mT scans, identify features that contribute to lesion detection accuracy, and explore super-resolution imaging at low-field. In this prospective, cross-sectional study, same-day brain MRI (FLAIR, T1w, and T2w) scans were collected from 36 adults (32 women; mean age, 50 ± 14 years) with known or suspected MS using Siemens 3T (FLAIR: 1 mm isotropic, T1w: 1 mm isotropic, and T2w: 0.34-0.5 × 0.34-0.5 × 3-5 mm) and Hyperfine 64mT (FLAIR: 1.6 × 1.6 × 5 mm, T1w: 1.5 × 1.5 × 5 mm, and T2w: 1.5 × 1.5 × 5 mm) scanners at two centers. Images were reviewed by neuroradiologists. MS lesions were measured manually and segmented using an automated algorithm. Statistical analyses assessed accuracy and variability of segmentations across scanners and systematic scanner biases in automated volumetric measurements. Lesions were identified on 64mT scans in 94% (31/33) of patients with confirmed MS. The average smallest lesions manually detected were 5.7 ± 1.3 mm in maximum diameter at 64mT vs 2.1 ± 0.6 mm at 3T, approaching the spatial resolution of the respective scanner sequences (3T: 1 mm, 64mT: 5 mm slice thickness). Automated lesion volume estimates were highly correlated between 3T and 64mT scans (r = 0.89, p < 0.001). Bland-Altman analysis identified bias in 64mT segmentations (mean = 1.6 ml, standard error = 5.2 ml, limits of agreement = -19.0-15.9 ml), which over-estimated low lesion volume and under-estimated high volume (r = 0.74, p < 0.001). Visual inspection revealed over-segmentation was driven venous hyperintensities on 64mT T2-FLAIR. Lesion size drove segmentation accuracy, with 93% of lesions > 1.0 ml and all lesions > 1.5 ml being detected. Using multi-acquisition volume averaging, we were able to generate 1.6 mm isotropic images on the 64mT device. Overall, our results demonstrate that in established MS, a portable 64mT MRI scanner can identify white matter lesions, and that automated estimates of total lesion volume correlate with measurements from 3T scans.

Keywords: Hyperfine; Low-field MRI; Multiple sclerosis; Point-of-care MRI; Portable MRI; White matter lesions.

Graphical abstract

Fig. 1

Flow chart of study participants. Abbreviations: multiple sclerosis (MS), clinically isolated syndrome (CIS), neuromyelitis optica (NMO).

Fig. 2

MS lesions on 3T and 64mT pulse sequences. Paired 3T (A) and 64mT (B) images from a 66-year-old woman with stable RRMS. Sequences include T1w (left), T2w (center), and T2-FLAIR (right). Images from both scanners show deep gray matter lesions and periventricular white matter lesions. Note also the superior sagittal sinus is hyperintense in the 64mT T2w and T2-FLAIR sequences, but not at 3T.

Fig. 3

Manual lesion size measurements and interrater reliability. Raters from each site independently measured the maximum diameter (Dmax) of the smallest lesion (Sm) and largest lesion (Lg) in 3T and 64mT imaging for all patients. (A) For the largest lesion measurements, there was no significant difference between raters at 3T (t = 1.3, p = 0.19) or 64mT (t = 1.2, p = 0.23); additionally, there was no difference between 3T and 64mT measurements (t = 0.04, p = 0.97). (B) For the smallest lesion measurements, there was a significant difference between raters for 3T measurements (t = 4.83, p < 0.001) although 64mT measurements were not significantly different (t = 1.67, p = 0.11); additionally, the diameter of the smallest lesion was significantly lower (t = 19.6, p < 0.001) when measured on 3T (mean 2.1 mm) compared to 64mT (mean 5.7 mm). (C) Across all lesions there was a strong correlation (r = 0.90, p < 0.001) between raters. There was significant intraclass correlation for the largest lesion at 3T (ICC = 0.77, CI = [0.58–0.88]), largest lesion at 64mT (ICC = 0.91, CI = [0.83–0.96]), smallest lesion at 3T (ICC = 0.62, CI = [0.12–0.83]), and smallest lesion at 64mT (ICC = 0.66, CI = [0.4–0.82]), indicating a high degree of agreement between rater measurements for both 3T and 64mT.

Fig. 4

Quantitative comparison of lesion visibility and image blurring at 3T and 64mT. (A) Lesion conspicuity measures the intensity of a lesion relative to background tissue. Lesion conspicuity for the largest lesion was measured on 3T (light blue) and 64mT (dark blue) T2-FLAIR images for 10 patients. There was no significant difference in conspicuity between the scanners (paired t-test, t = 0.14, p = 0.89). (B) SNR compares mean lesion intensity to background noise. SNR was significantly higher in 3T images in the subset of 10 patients (paired t-test, t = 4.36, p = 0.00184). (C) Similarly, CNR, which compares the contrast between lesion and white matter to background noise, was also significantly higher in 3T imaging (paired t-test, t = 4.89, p < 0.001). (D-F) The variance of the Laplacian is a measure of image focus, with larger values indicating clearer images. We registered and resliced 3T to 64mT images and calculated this focus feature for both images. This process was carried out for (D) T2-FLAIR, (E) T1w, and (F) T2w sequences. Four subjects without T2w sequence pairs were excluded from panel D. For all sequences, low-field images were significantly more blurred than their resliced 3T counterparts (paired t-tests, T2-FLAIR: t = 11.6, p < 0.001, T1w: t = 9.5, p < 0.001, T2w: t = 19.5, p < 0.001). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5

Automated lesion segmentations at 3T and 64mT overlap. (A) 64mT FLAIR images for three cases (left) with automated lesion segmentations generated from the 64mT images using MIMoSA overlaid (right). (B) Corresponding 3T FLAIR images for the same three cases (left) with 3T based segmentations (right). Patients from top to bottom are a 51-year-old female with RRMS, 44-year-old female with RRMS, and 71-year-old female with RRMS. All images were coregistered to 64mT T1-weighted images for comparison. Segmentations generated from 64mT and 3T scanners show similar patterns, although examples of false-positive segmentation in the sagittal sinus at 64mT can be seen in the top and bottom patients.

Fig. 6

Total lesion volume measured at 3T and 64mT shows agreement. (A) 3T and 64mT total lesion volume estimates were strongly correlated (Pearson’s correlation, r = 0.89, p < 0.001). However, when compared to y = x there is a clear bias towards over-segmentation at low levels of total lesion volume. (B) A Bland-Altman plot illustrates the level of agreement between 3T and 64mT segmentation volumes (bias −1.6 ml, standard error of measurement = 5.2 ml, 95% limit of agreement −19.0 to 15.9 ml). The Pearson’s correlation (r = 0.74, p < 0.001) in dark blue further indicates over-segmentation at 64mT when lesion volume is low and under-segmentation when lesion volume is high. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 7

Lesion size and intensity influence detection rate. (A) The detection rate, or true positive rate (TPR), steadily increases with lesion size, with 93% detected at > 1 ml, and all lesions >1.5 ml being detected. The false discovery rate (FDR) decreases with lesion size, with 36% false discovery rate at > 1 ml, 22% at > 1.5 ml, and 3% at > 2.5 ml. Though the x-axis was limited to 4 ml for illustrative purposes, lesions > 20 ml were found in the dataset. (B) To analyze the relationship between lesion intensity and detection rate, image intensity values were first normalized using White Stripe (Shinohara et al., 2014). While detection rate increases as mean lesion intensity increases, the FDR remains high (>75%) across lesion intensities. The high number of false positive detections was driven by hyperintense veins and peripheral signal artifacts, as seen in Fig. S3.

Fig. 8

Multi acquisition image averaging can increase lesion conspicuity and resolution. This figure depicts a 3x4x5 mm (0.06 ml) subcortical left frontal white matter lesion in a 53-year-old woman with stable RRMS and compares 64mT FLAIR images generated from multi-acquisition image averaging to 3T imaging. The lesion is readily apparent on 3T imaging (A); however, it could not be discerned in a single 64mT acquisition (D). Volume averaging of multiple acquisitions with repositioning between scans did reveal the lesion on the low-field system (B & C). The lesion was discernible for N ≥ 3 multi acquisition averages. The lesion was manually segmented on 3T, and the ratio of mean lesion intensity to ipsilateral adjacent white matter (WM) is given as an estimate of lesion conspicuity (red dot). In 64mT imaging, the ratio steadily increases with additional acquisition averages (blue dots). With 8 vol averages, there was a 53% increase in lesion conspicuity. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 9

Super-resolution images generated from orthogonal slice directions. This figure demonstrates a super-resolution approach using anisotropic image acquired in orthogonal slice directions (axial, sagittal, and coronal). The patient is a 69-year-old man with stable RRMS. (A) 3T FLAIR imaging demonstrates a right periventricular lesion (conspicuity = 0.18) in the three orthogonal planes. (B) 64mT imaging of the same lesion (conspicuity = 0.11) using an axial FLAIR acquisition with sagittal and coronal reformatted images. (C) Corresponding axial, sagittal, and coronal slices from separate acquisitions in each direction (conspicuity: axial = 0.11, sagittal = 0.09, and coronal = 0.13) registered to the axial image by affine transformation. (D) Corresponding super-resolution images generated by averaging the coregistered axial, sagittal, and coronal acquisitions in C (conspicuity: linear = 0.11, nearest-neighbor = 0.13).