Bedside detection of intracranial midline shift using portable magnetic resonance imaging

Neuroimaging is crucial for assessing mass effect in brain-injured patients. Transport to an imaging suite, however, is challenging for critically ill patients. We evaluated the use of a low magnetic field, portable MRI (pMRI) for assessing midline shift (MLS). In this observational study, 0.064 T pMRI exams were performed on stroke patients admitted to the neuroscience intensive care unit at Yale New Haven Hospital. Dichotomous (present or absent) and continuous MLS measurements were obtained on pMRI exams and locally available and accessible standard-of-care imaging exams (CT or MRI). We evaluated the agreement between pMRI and standard-of-care measurements. Additionally, we assessed the relationship between pMRI-based MLS and functional outcome (modified Rankin Scale). A total of 102 patients were included in the final study (48 ischemic stroke; 54 intracranial hemorrhage). There was significant concordance between pMRI and standard-of-care measurements (dichotomous, κ = 0.87; continuous, ICC = 0.94). Low-field pMRI identified MLS with a sensitivity of 0.93 and specificity of 0.96. Moreover, pMRI MLS assessments predicted poor clinical outcome at discharge (dichotomous: adjusted OR 7.98, 95% CI 2.07–40.04, p = 0.005; continuous: adjusted OR 1.59, 95% CI 1.11–2.49, p = 0.021). Low-field pMRI may serve as a valuable bedside tool for detecting mass effect.

Technical and imaging parameters. We used a 0.064 T MRI system (Hyperfine, Guilford, CT, USA) to obtain pMRI exams at the patient's bedside (Fig. 1). The pMRI device has a height of 140 cm and a width of 86 cm. The device contains an 8-channel head coil, which has a height of 26 cm and width of 20 cm. The vertical and horizontal clearance of the pMRI are 32 cm and 55 cm, respectively. The scanner uses a biplanar 3-axis gradient system with a peak amplitude of 26 mT/m (on Z-axis) and 25 mT/m (on X-and Y-axis), operates from a standard 110 V, 15A electrical outlet, and does not require any cryogens. All pMRI exams were conducted in single-patient ICU rooms, which included the presence of nearby ferromagnetic equipment (e.g., vital signs monitors, intravenous infusion pumps, ventilators, compressed gas cylinders, and dialysis machines).
The pMRI device is capable of obtaining T2-weighted, T1-weighted, FLAIR, and diffusion-weighted imaging pulse sequences. Image sequences were selected through an electronic interface (iPad Pro third generation). www.nature.com/scientificreports/ Each acquired sequence was displayed on the iPad in real-time throughout image acquisition and processing. All pMRI images were automatically uploaded in DICOM format to a cloud-based server upon completion of the pMRI exam.
Each pMRI scanner must meet factory imaging performance criteria prior to its delivery to a clinical site. These criteria entail the scanning of an image quality phantom to ensure the scanner fulfills performance metrics established by the National Electrical Manufacturers Association (NEMA). System quality assurance results, including metrics on geometric distortion (see Supplementary Material online), are reported to the U.S. Food and Drug Administration. Additionally, each pMRI scanner arrives with an image quality phantom. The phantom is scanned each month on-site, and the phantom images are uploaded to the Hyperfine Cloud Picture Archive and Communication System, which allows for monitoring and evaluation of these phantom images for calibration and quality assurance purposes. Imaging analysis. MLS  Following previously published approaches in stroke populations 10,30,31 , MLS was defined as any deviation of the septum pellucidum from the midline. MLS was assessed as a continuous and dichotomous (present or absent) variable. Continuous MLS measurements were obtained by drawing a line from the anterior and posterior attachments of the falx cerebri and then drawing a second, perpendicular line to the septum pellucidum at the point of maximal deviation. MLS was measured as the length in millimeters of the second line (continuous variable) (Fig. 2) 11,[30][31][32][33] . MLS greater than 2 mm is associated with poor clinical outcome 34,35 , so we defined any rater's MLS measurement ≥ 2 mm to indicate the presence of significant MLS (dichotomous variable). For both pMRI and SOC MLS measurements, we generated a consensus (dichotomous) and averaged (continuous) MLS assessment for each patient. The consensus (present or absent) MLS assessment was obtained from the majority consensus of the three raters' dichotomous MLS assessments. The averaged MLS assessment was obtained by averaging the three raters' continuous MLS measurements.
Neurological outcome. Functional outcome at discharge was assessed by the modified Rankin Scale (mRS). The mRS scale ranges from 0 (no residual stroke symptoms) to 6 (death), and mRS scores were dichotomized into good (0-3) and poor clinical outcomes (4-6) 36 .

Statistics.
We present categorical variables as numbers (%) and continuous variables as mean (standard deviation [SD]) or median (interquartile range [IQR]), as appropriate. Interrater reliability between the raters' pMRI and SOC MLS assessments was computed using the Fleiss kappa (κ) statistic for dichotomous MLS assessments and the intraclass correlation coefficient (ICC) for continuous MLS measurements.
Low-field pMRI-based MLS measurements were compared to SOC-based measurements, which were considered the ground truth. We first compared each raters' individual measurements (κ for dichotomous, ICC for continuous). We then compared majority consensus and averaged MLS measurements between pMRI and SOC imaging studies (κ for dichotomous, ICC for continuous). The agreement between pMRI and SOC MLS assessments was also studied by the Bland-Altman method with calculation of bias and limits of agreement. To account for confounding effects due to continuous development of the pMRI system, identical analyses were performed in three groups of patients: patients scanned using software versions RC3 and RC4, patients scanned using software versions RC5 and RC6, and patients scanned using software versions RC7 and RC8.
To assess the relationship between MLS and discharge functional outcome, we performed a χ 2 test to see if there was an association between the presence of MLS and discharge mRS scores. We then assessed the relationship between dichotomous and continuous MLS assessments and clinical outcomes using unadjusted and adjusted binary logistic regression models. We expressed the effect of MLS on functional outcome as unadjusted and adjusted common odds ratios (cOR and acOR, respectively). In stratified analyses, we evaluated ischemic stroke (IS) and intracranial hemorrhage (ICH) patients separately. To adjust forseline prognostic variables, our adjusted models included sex, race, age, stroke severity (NIH Stroke Scale score at admission), history of diabetes mellitus, atrial fibrillation, and prior stroke. All statistical analyses were performed using RStudio version 1 We recorded pMRI examination times for five non-intubated stroke patients; these recorded times are also noted in a different report 29 . Single-sequence (T2-weighted or FLAIR) pMRI exams were obtained in 21:00 ± 0:10 [SD] minutes. Point-of-care pMRI exams required 8:33 ± 0:09 min for set-up, which entailed bringing the scanner into the room, positioning the pMRI behind the patient's bed, and boosting the patient into the scanner. T2-weighted and FLAIR sequences were acquired in 7:01 ± 0:06 and 8:45 ± 0:03 min, respectively. After the imaging protocol was completed, removing the patient from the scanner and restoring the patient's room to the prior state required 4:27 ± 0:03 min.
To account for confounding effects due to evolving improvements of the pMRI system, the abovementioned analyses were performed in three groups of patients: patients scanned using pMRI software versions RC3 and RC4, patients scanned using pMRI software versions RC5 and RC6, and patients scanned using pMRI software versions RC7 and RC8.
For patients scanned using software versions RC3 and RC4 (n = 26), the agreement between pMRI-measured and SOC-measured MLS assessments was k = 0.75, p = 1.3 × 10 -4 and ICC = 0.73, p = 6.3 × 10 -4 ( Table 2). The Bland-Altman plot of averaged pMRI and SOC MLS assessments showed a bias of − 0.40 mm and limits of agreement from 1.90 mm to − 2.70 mm (Fig. 3b). For this group of patients, low-field pMRI detected the presence of significant MLS with a sensitivity of 0.80 and specificity of 0.95.
For patients scanned using software versions RC5 and RC6 (n = 28), the agreement between pMRI-measured and SOC-measured MLS assessments was k = 0.90, p = 1.7 × 10 -6 and ICC = 0.97, p = 6.4 × 10 -14 ( Table 2). The Bland-Altman plot of averaged pMRI and SOC MLS assessments showed a bias of 0.01 mm and limits of  www.nature.com/scientificreports/ agreement from 1.39 mm to − 1.37 mm (Fig. 3c). For this group of patients, low-field pMRI detected the presence of significant MLS with a sensitivity of 1.00 and specificity of 0.96. For patients scanned using software versions RC7 and RC8 (n = 12), the agreement between pMRI-measured and SOC-measured MLS assessments was k = 1.00, p = p = 5.3 × 10 -4 and ICC = 1.00, p = 1.6 × 10 -12 ( Table 2). The Bland-Altman plot of averaged pMRI and SOC MLS assessments demonstrated a bias of 0.05 mm and limits of agreement from 0.59 mm to − 0.49 mm (Fig. 3d). For this group of patients, low-field pMRI detected the presence of significant MLS with a sensitivity of 1.00 and specificity of 1.00.
There was substantial agreement of pMRI-based and SOC-based MLS assessments for all software versions, but continuous improvement in pMRI image quality (Fig. 4) corresponded with increased diagnostic capability of the pMRI to accurately detect and measure midline shift.

Discussion
We report the use of low-field pMRI for bedside assessment of MLS in patients with IS and ICH. This approach enabled the acquisition of bedside neuroimaging exams that visualized MLS, a well-known marker of mass effect and cerebral edema 5,10,11,37 . We show that MLS measurements on pMRI images are consistent with measurements obtained on conventional MRI and CT studies. We also demonstrate that MLS on pMRI neuroimaging is associated with worse discharge functional outcome, recapitulating a well-established clinical relationship [5][6][7][8][9][10][11] . Neuroimaging studies are integral to the initial assessment and neurological monitoring of patients with acute brain injuries. In conventional imaging pathways, patients must be transported to a dedicated imaging suite. However, intrahospital transport of patients is associated with numerous cardiovascular and respiratory risks [23][24][25][26][27] , which may render the acquisition of conventional CT or MRI imaging unfeasible for clinically unstable patients 38 . MRI scanners operating at low-field magnetic strength enable scanning outside of traditional imaging suites, as they are compatible with nearby ferromagnetic material. While previous approaches in low-field MRI, such as pre-polarized MRI, have been explored 39 , there has not been a low-field MRI device for head imaging that is entirely portable and has been successfully deployed in a clinical environment.
We previously reported the first use of a highly mobile low-field pMRI device to obtain head imaging at the bedside of intensive care patients 28 and to evaluate intracerebral hemorrhage 29 . The current study extends our understanding of the unique applications of pMRI in evaluating neuropathology at the bedside. In braininjured patients, attribution of a change in the level of arousal often requires neuroimaging to diagnose MLS, a well-known marker of mass effect and brain-swelling 5,10,11,37 . MLS is one of multiple important biomarkers for acute brain injuries, and the detection of significant MLS can serve as a radiologic indicator for treatment with hyperosmolar agents or neurosurgical interventions, such as decompressive craniectomy and hematoma evacuation. Monitoring changes in MLS is also important when evaluating the efficacy of such treatments, as unresolved MLS predicts worse clinical outcome [40][41][42] while reversal of MLS is associated with improved consciousness and survival [43][44][45] . Our data show that pMRI can identify and quantify MLS with clinically significant accuracy, demonstrating the unique utility of low-field pMRI as a bedside tool for monitoring MLS. Further study is required to assess the ability of pMRI to detect smaller morphological markers of mass effect, includingsal cistern compression, ventricle effacement, and brainstem displacement. www.nature.com/scientificreports/ Our study has several limitations. First, pMRI and conventional imaging studies were not obtained simultaneously, with an average time difference of 13 ± 8 [SD] hours. Since MLS is a dynamic neurological marker, this limitation may have induced discrepancies between the pMRI and conventional MLS assessments. Second, it is important to note that patients under 18 years of age and those with a cardiovascular implantable device were not included in this study, so these results cannot be extrapolated to those populations. However, given recent reports of safe, feasible MRI at 1.5 T for patients with cardiovascular implants 46-49 , using a low magnetic field pMRI in this patient population is theoretically possible and requires further study. Finally, patients were imaged at a single-center ICU. Replication of these findings at multiple centers and in clinical environments outside an intensive care setting (e.g., emergency medicine) is necessary before generalizing the results of the current study.
Our approach has several unique aspects. First, we successfully deployed an innovative MRI technology that enabled neuroimaging at the bedside. Many brain pathologies, like MLS, evolve over a dynamic time window and, in turn, require serial imaging. Similarly, neurosurgical interventions often require preoperative and  www.nature.com/scientificreports/ postoperative imaging. Repeated transport of critically ill patients to neuroimaging suites may be unfeasible and hazardous. In a reversal of the current imaging paradigm, we deployed a pMRI directly to the bedside of stroke patients and acquired whole-brain imaging that detected MLS within 10 min (7:01 min for T2W, 8:45 min for  FLAIR). Point-of-care pMRI can serve as a safe and viable approach to neuroimaging when serial transport to conventional imaging suites is otherwise contraindicated. Additionally, pMRI operates on a low-field magnetic strength. Traditional MRI systems operate on high-field magnetic strength, requiring rigid safety precautions each time a healthcare worker enters an MRI suite. These constraints, in addition to the enclosed design of traditional MRI scanners, prevent healthcare workers from easily accessing and caring for patients during conventional MRI exams. In our study, the low-field magnetic strength of the pMRI allowed nurses to freely enter and exit the patient's ICU room during bedside exams without projectile risk. Moreover, the open geometry design of the pMRI enabled nurses to directly contact and care for the patient (e.g., temperature monitoring, intravenous injection of medication) throughout the imaging exam. These results outline the potential use of pMRI for intensive care patients that require frequent attention and care.
It is important to contextualize the use of low-field pMRI with other portable imaging techniques, including portable CT (pCT) and transcranial ultrasound imaging. Both pCT and ultrasound imaging have been used to detect MLS of the brain at the bedside 50,51 . Ultrasound imaging is an accessible bedside technique that is best suited for monitoring cerebral blood flow and vessel imaging. However, ultrasound's capacity for structural imaging is limited by the distortion of ultrasound beams as they cross the skull 52 . Point-of-care pCT is a well-explored imaging modality that can provide valuable imaging at the bedside 53 . Point-of-care pCT scanners are capable of non-contrast CT, CT angiography, and CT perfusion, which enables the detection of both anatomical lesions, including hemorrhage 54 and subacute ischemia 55 , and vascular abnormalities, such as an ischemic penumbra 56 and large-vessel occlusions 57 . Similar to pCT scanners, the low-field pMRI evaluated in this report can detect hemorrhage and ischemia. The low-field pMRI device can also detect restricted diffusion through diffusionweighted imaging 28 . However, the low-field pMRI device does not currently have MR angiography or perfusion weighted imaging, limiting its ability to detect ischemic penumbras and large-vessel occlusions.
Low-field pMRI and pCT are the two most analogous portable imaging techniques, but there are several limitations to bedside pCT which have prevented its widespread adoption in dynamic hospital settings. Compared to fixed CT scanners, pCT devices have lower spatial resolution, amplified noise, and higher radiation risks [58][59][60] . Moreover, bedside pCT requires highly trained technicians and lead shielding around the point-of-care, limiting its ease of use. In contrast, low-field pMRI does not use any ionizing radiation nor require specialized MRI technicians for use. In this study, all pMRI exams were conducted by research assistants under the supervision of nearby nurses. Moreover, pMRI exams were configured by simply connecting an iPad to the pMRI's local hotspot and selecting a pre-configured imaging protocol on a user interface hosted on a web browser, demonstrating pMRI's ease of use. Finally, pCT captures only one type of structural image, while pMRI can obtain multiple imaging sequences, including T2-weighted, FLAIR, and diffusion-weighted imaging 28 . Compared to pCT, the unique strengths of pMRI lie within its capacity to obtain multimodal imaging in a safe and feasible manner.
Our study demonstrates the use of a highly mobile low-field MRI scanner to detect clinically significant MLS at the patient's bedside. Future studies will need to delineate the strengths and limitations of pMRI scanning for different timepoints of brain injury, patient populations, and medical environments. Further validation of the pMRI's sensitivity and specificity to other intracranial pathologies, such as ischemic stroke and intracranial hemorrhage, is also required. Nonetheless, the current results demonstrate the clinical feasibility of pMRI in a complex medical environment 28 , and we report the first assessment of MLS as a surrogate of mass effect using a portable, bedside MRI device. In instances where single or repeated transport of patients to conventional imaging studies is unfeasible, point-of-care pMRI may serve as a valuable bedside tool that can facilitate the study of disease processes over a dynamic profile. www.nature.com/scientificreports/