Portable, bedside, low-field magnetic resonance imaging for evaluation of intracerebral hemorrhage

Radiological examination of the brain is a critical determinant of stroke care pathways. Accessible neuroimaging is essential to detect the presence of intracerebral hemorrhage (ICH). Conventional magnetic resonance imaging (MRI) operates at high magnetic field strength (1.5–3 T), which requires an access-controlled environment, rendering MRI often inaccessible. We demonstrate the use of a low-field MRI (0.064 T) for ICH evaluation. Patients were imaged using conventional neuroimaging (non-contrast computerized tomography (CT) or 1.5/3 T MRI) and portable MRI (pMRI) at Yale New Haven Hospital from July 2018 to November 2020. Two board-certified neuroradiologists evaluated a total of 144 pMRI examinations (56 ICH, 48 acute ischemic stroke, 40 healthy controls) and one ICH imaging core lab researcher reviewed the cases of disagreement. Raters correctly detected ICH in 45 of 56 cases (80.4% sensitivity, 95%CI: [0.68–0.90]). Blood-negative cases were correctly identified in 85 of 88 cases (96.6% specificity, 95%CI: [0.90–0.99]). Manually segmented hematoma volumes and ABC/2 estimated volumes on pMRI correlate with conventional imaging volumes (ICC = 0.955, p = 1.69e-30 and ICC = 0.875, p = 1.66e-8, respectively). Hematoma volumes measured on pMRI correlate with NIH stroke scale (NIHSS) and clinical outcome (mRS) at discharge for manual and ABC/2 volumes. Low-field pMRI may be useful in bringing advanced MRI technology to resource-limited settings.

T imely and accessible neuroimaging is a critical step in the diagnostic workup of patients presenting with suspected acute brain injury such as stroke 1,2 . Since intracerebral hemorrhage (ICH) is a contraindication for thrombolytic therapy 3,4 , ruling out the presence of blood is one of the main decision steps in acute stroke care. Current guidelines for the early management of stroke from the American Heart Association (AHA) advise that all patients receive rapid brain imaging on hospital arrival prior to initiating any thrombolytic treatment 5 . Non-contrast computed tomography (CT) of the head has historically been the imaging modality of choice for diagnosing ICH due to its convenience and high sensitivity for hemorrhage [6][7][8] . However, a growing body of recent evidence has demonstrated that multimodal magnetic resonance imaging (MRI) is as accurate as CT for detecting acute brain hemorrhage [9][10][11][12][13][14][15][16][17][18] and avoids the radiation exposure associated with CT 19 . Certain strategies have previously been developed to reduce CT radiation burden 20,21 . Nevertheless, studies comparing CT and MRI demonstrate that magnetic resonance technology has higher sensitivity to ischemia, leukoencephalopathy, and classifying forms of extra-axial hemorrhage 22,23 . Furthermore, MRI is shown to offer more precise anatomic depiction of neuropathology and sharper resolution of soft tissue and contrast in comparison to CT 24,25 .
In addition to acute stroke evaluation, other clinical contexts, such as post-neurosurgical assessment of patients, require neuroimaging evaluation to detect the presence of ICH. Neuroimaging is also essential for characterizing ICH, which aids in diagnosing the etiology of ICH, clinical management, and prognosis formation. For instance, non-lobar ICH is often caused by hypertension, and intraventricular and cerebellar hemorrhage require neurosurgical intervention for cerebrospinal diversion or suboccipital decompression. Additionally, clinicians commonly use ICH volume as a critical determinant of prognostication 26 .
Traditionally, neuroimaging requires patient transport to a centralized dedicated radiology suite, which is costly in both time and resources [27][28][29][30] . Conventional MRI systems operate at high magnetic field strengths (1.5-3T) 31 , which require specialized infrastructure, highly trained technicians, and rigid safety precautions 27,32,33 . As a result, MRI is not easily accessible for unstable patients or for populations in resource-limited settings where secure access radiology suites are not available throughout the day 34 . In ICH patients being transported for neuroimaging, potential adverse events include increased intracranial pressure, cardiovascular instability, and compromise of monitoring equipment and intravenous lines 35,36 . Recent advances in lowfield MRI (<0.2T) have allowed for imaging outside of strict access-controlled radiology suites and in the presence of ferromagnetic materials at the point-of-care 27,[37][38][39][40][41] . The ability to operate at low magnetic field strength eliminates the need for expensive superconducting magnets, can result in fewer susceptibility artifacts, and offers increased flexibility in open geometry design and improved T 1 contrast 27,38,40 . Previously, midfield MRI technology (0.2-1T) has been employed for the acquisition of clinically useful imaging in critical care units 40,41 and the use of low-field MRI technology has been posited as a meaningful solution for stroke 25 . However, these efforts were based on large, fixed imaging systems rather than mobile, bedside units. A prior report presented a mobile and efficient low-field (23 mT) MRI prototype for neonatal applications, however no patient imaging was performed 37 .
We report the use of a low-field (0.064T), portable MRI (pMRI) system ( Fig. 1) in critically ill patients presenting with ICH. Our primary objective was to demonstrate the ability to deploy pMRI neuroimaging at the hospital bedside and provide initial evaluation for detection of ICH. We provide a systematic assessment of ICH detection using neuroimaging derived from pMRI. Specifically, we report the sensitivity and specificity of ICH detection and the accuracy of ICH localization. In addition, we explore the association between pMRI-derived ICH characteristics and clinical outcome.

Results
Study Cohort and Safety. We obtained 119 pMRI examinations on 104 patients presenting to the neuroscience intensive care unit (NICU) or emergency department (ED) with a confirmed diagnosis of ICH or acute ischemic stroke (AIS). Eleven patients were imaged at serial timepoints. Nine exams were excluded from this analysis due to the patient having a body habitus that prevented full brain insertion into the scanner's head coil and produced an incomplete field-of-view. An additional six exams were significantly degraded due to patient motion and were excluded.  [11] years) with a diagnosis of no intracranial abnormality scanned at YNHH. Two healthy controls were serially imaged. Of the total cohort, 5 pMRI examinations reached the level of the midbrain (4%), 25 exams reached the level of the pons (17%), 107 exams reached the level of the medulla (74%), and 7 exams reached the level of the lateral ventricles (5%). Table 1 summarizes the demographic and clinical characteristics of the study population.
No adverse events occurred. Patients remained connected to all intravenous lines and ICU monitoring equipment during sequence acquisition. Table 2 delineates the time required for pMRI imaging compared to conventional MRI. The mean examination time was 17:51 min for a pMRI protocol that included pre-scan calibration, localizer, T2W, and FLAIR imaging. The static magnetic field, gradient, and RF pulses of the pMRI scanner did not interfere with the operation of infusion pumps, mechanical ventilators, or hemodialysis machines. Research associates and clinical staff were able to remain in the room during sequence acquisition.
Sensitivity and specificity rater evaluation. Neuroradiologists were provided a total of 144 exams ( Fig. 2, compared to traditional high-field MRI or CT. Figure 3 demonstrates false negative neuroimaging findings for ICH. To account for confounding effects due to evolving improvements in scanner software and hardware, an identical analysis to above was performed in two subsets by grouping exams by software versions into the first half of the study and its respective scanner software versions (RC3, RC4, RC5), and the second half of the study and its respective scanner software versions (RC6, RC7, RC8).
Hematoma volume and localization. Low-field pMRI examinations were obtained for 56 ICH patients. Fourteen exams were excluded from this sub-analysis since they did not have a conventional imaging study (CT or 1.5/3T MRI) within 36 h of the pMRI examination. As a result, 42 ICH patient exams were provided to one rater for manual segmentation. Five T2W and two FLAIR pMRI cases were excluded since their corresponding conventional sequence was not obtained; an additional two FLAIR pMRI cases on the earliest software version were excluded since the lesion was not visualized to segment. Manually segmented lesions on pMRI T2W and FLAIR sequences had a median [IQR] of 5.58 cc [3.38-16.9] and 8.61 cc [3.25-28.4], respectively. To estimate hematoma volume using the ABC/2 method 44 , 40 of the 42 ICH patient exams were provided to four raters for evaluation; two lesions with non-ellipsoid morphology were excluded due to the inaccuracy of the ABC/2 measurement in these cases 45 and two FLAIR pMRI on cases the earliest software version were excluded since the lesion was not visualized. Hematoma volumes estimated using the ABC/2 method on pMRI T2W and FLAIR sequences were averaged across raters and had a median [ (Fig. 4a1). Bland-Altman  Table S1 and plots shown in Fig. S1.  (Table 3).

Discussion
We report the validation of portable MRI (pMRI) in evaluating intracerebral hemorrhage (ICH). These results demonstrate the successful deployment of a low-field pMRI device to the bedside of critically ill patients with ICH. With this approach, we obtained neuroimaging results that enabled detection and characterization of ICH. Our observation that ICH volume measured on pMRI is associated with both stroke severity and patient outcome further validates this approach as ICH volume is a wellestablished predictor of outcome 46 . These results suggest that  (a1, b1). 95% confidence intervals are represented by bands (a1, b1, c1) and dashed gray lines (a2-3, b2-3, c2-3). Neuroimaging is the most reliable method of differentiating ICH from non-ICH etiologies of focal neurological deficit. Observational series two decades earlier provided some of the first evidence demonstrating at least equivalent sensitivity for MRI in detecting ICH compared to CT 9 . As many as 70% of patients undergo MRI during a hospital admission. However, the timing and availability of MRI evaluation is routinely limited by patient transport, access to secure access facilities, and safety concerns, either due to an individual's clinical status or potential danger in a high magnetic field strength environment. The risks inherent in intrahospital transport to dedicated radiology suites (CT or 1.5/3T MRI) are reported between 20-70%, even under the supervision of a well-trained transport team 47,48 . While portable CT scanners have been introduced to address these concerns 29,30,35,47,[49][50][51][52] , the lower spatial resolution, amplified noise, and higher radiation dose compared to their fixedlocation counterpart render them disadvantageous to broad adoption in a dynamic hospital setting 47,50,52 . Moreover, the portable CT scanners require trained technicians and lead shielding to operate 49 . Conversely, the low-field pMRI solution can be used in the absence of a specialized MRI technician and requires minimal training of operator staff. Low-field pMRI does not have a liquid cryogenic requirement or use ionizing radiation, permits unrestricted access into the hospital room during acquisition, and allows for the added benefits of MRI evaluation compared to CT [22][23][24][25] .
Using a pMRI approach, we demonstrate that ICH can be detected at the bedside using low-field (0.064T) magnetic resonance technology. This approach allows for a reversal in the clinical paradigm, wherein the pMRI comes to the patient. Our results extend our preliminary success in deploying a pMRI solution to the bedside of critically ill patient populations 53 . Due to the low magnetic field strength of the device, hospital staff and patients were able to safely enter the clinical environment, with no need to remove ferromagnetic objects required for clinical care. The scanner's open geometry design allowed for easy access to intravenous lines, ventilation tubing, and intraventricular drains during pMRI examinations. Patients and research staff did not experience any adverse events during pMRI deployment and could safely remain in the hospital room during scan acquisition.
Our objective was to characterize the presence of ICH in a braininjured population focusing on characteristics that are known to have clinical relevance, including detection of ICH, description of anatomical location, and hematoma volume. The overall sensitivity for ICH was notable, especially for supratentorial ICH, which is the most common presentation 42,43 . However, because detection and accurate measurement of ICH is essential for acute stroke care and hemorrhage in the posterior fossa is the more life-threatening subtype 54 , further improvements in pMRI hardware and software will be required. Indeed, sensitivity improvements were observed even during this study as upgraded hardware and software versions became available. Furthermore, our results demonstrate that hematoma volumes measured on pMRI are in agreement with those measured on conventional imaging. To further validate this volumetric approach, the Euclidean distances between the centroids of the manually segmented hematomas and the centroids of the point ROIs at four anatomical locations were compared across modalities. We observed a strong correlation between distances across modalities, demonstrating congruence of lesion localization on pMRI with conventional imaging. Overall, further study is required to determine how detection can be most efficiently improved with hardware and software advances, image reconstruction tools, automated detection strategies, and reader training.
Exams from six patients were not interpretable due to motion artifacts and degraded images. The low field strength of the magnet and the resulting lower signal-to-noise ratio makes these images particularly susceptible to patient motion. Domain transform manifold learning is one of several post-processing techniques in the literature that we speculate could be applied to pMRI reconstruction to mitigate this limitation in the future 55 . Future clinical utility will require further quality improvements in motion-related artifact. Furthermore, exams from nine patients were excluded from this analysis due to the patient having a body habitus that prevented full entry into the scanner's opening and produced a limited field-of-view. While this small subset of patients obtained a pMRI exam with a field-of-view that did not capture the infratentorial pathology, the majority of patients had a complete pMRI field-of-view that reached to the level of the medulla and pons, allowing for full brain insertion and infratentorial evaluation.
It is also important to note the patient populations that our study did not fully evaluate. Low-field pMRI examination largely occurred within the subacute phase and future studies will need to evaluate ICH in the hyperacute setting, given differences in ICH appearance on MRI at different time points 56,57 . In addition, our study did not evaluate for the presence of subarachnoid hemorrhage. Along with improved detection of ICH in infratentorial compartments and evaluation of subarachnoid hemorrhage, the results in the current study should be confirmed in larger multicenter studies that investigate all forms of intracranial hemorrhage. In the future, it would be beneficial to conduct a systematic study to help us understand which populations can and cannot receive pMRI scans across a variety of applications.
There are several strengths in the current analysis. First, we were able to integrate into the clinical workflow an innovative Hematoma volume measurements derived from pMRI exams were associated with impaired cognitive status and worse discharge functional outcome in unadjusted and adjusted multivariable linear regression models. Adjusted linear regression models included sex, race, and age. pMRI portable magnetic resonance imaging, T2W T2-weighted, FLAIR fluid-attenuated inversion recovery, NIHSS National Institutes of Health Stroke Scale, mRS Modified Rankin Scale.
solution that facilitated bedside detection of ICH using MRI technology. Second, we used several blinded raters, all with established expertize reading clinical neuroimaging studies but with a range of experience, supporting generalizability of the clinical interpretation. In using a large, systematic dataset we are able to report both the proof-of-concept as well as the limitations of detection in a single-center cohort, acknowledging that point estimates for detection will require further validation in large cohorts across multiple care environments. Moreover, demonstrating an established relationship between pMRI-ascertained ICH volume with cognitive impairment and patient outcome provides additional support of pMRI as a bedside neuroimaging solution. We also report initial signal intensity characteristics of the lesions seen on T2W and FLAIR sequences using pMRI, which have been used to interrogate mechanisms of injury. Lastly, we were able to obtain serial imaging with the pMRI in this study, suggesting that pMRI may be used to understand dynamic changes in ICH pathology over time, which is a structural limitation in current MRI approaches. These results demonstrate that pMRI is a safe and feasible neuroimaging solution that contributes valuable information in ICH evaluation. Overall, we report a systematic assessment of ICH using a portable MRI device that can be integrated into the clinical workflow to provide timely diagnostic neuroimaging at the bedside. Further study is required in prospective multicenter studies.

Methods
Setting, participants, and study design. This observational study was performed at Yale New Haven Hospital (YNHH) in New Haven, Connecticut from July 2018 to November 2020 under an Institutional Review Board (IRB) protocol approved by Yale Human Research Protection Program. From July 2018 to March 2020, the study was conducted with an investigational device exemption (IDE) and informed consent was obtained from all patients or their legally authorized representative. Starting in March 2020, pMRI neuroimaging examinations were obtained as part of the patients' clinical care under U.S. Food and Drug Administration (FDA) general clearance for portable imaging systems during the COVID-19 public health emergency 58 . The pMRI scanner later received specific FDA approval on 11 August 2020, and pMRI examinations continued as part of the patients' clinical care. This study complied to all relevant ethical regulations as outlined by the Yale Human Research Protection Program and U.S. FDA. Individuals who presented with a focal neurological deficit and were admitted to the emergency department (ED) or neuroscience intensive care unit (NICU) were screened for eligibility. Participant eligibility was determined based on admitting diagnosis, clinical exam, and baseline standard-of-care imaging during hospitalization. Bedside pMRI examinations were not obtained on patients with the presence of at least one of the following contraindications to conventional MRI evaluation: cardiac pacemakers or defibrillators, implantable drug pumps, deep brain stimulators, vagus nerve stimulators or cochlear implants, pregnancy, and cardiorespiratory instability. Patients who received a pMRI examination with the following inclusion criteria were included in this analysis as stroke patients 1 : a clinical diagnosis of intracerebral hemorrhage (ICH) or acute ischemic stroke (AIS) confirmed on the radiological report of the closest conventional imaging study (1.5/3T MRI or CT) 2 ; the acquisition of both T2-weighted (T2W) and Fluid-Attenuated Inversion Recovery (FLAIR) pMRI sequences; and 3 a patient body habitus that permitted positioning for full brain imaging within the scanner's head coil. Patients who received a pMRI examination with the following inclusion criteria were included in this analysis as healthy controls 1 : no past medical history of neuropathology; 2 a confirmed diagnosis of no intracranial abnormality on all available standard-ofcare imaging reports; 3 both T2W and FLAIR pMRI sequences obtained; and 4 a patient body habitus that permitted positioning for full brain imaging within the scanner's head coil. Demographic characteristics, medical history, baseline functional independence, clinical presentation, discharge functional outcome, and standard-of-care hospital imaging findings were collected for each subject from the electronic medical record.
In addition, non-patient healthy controls were recruited at Hyperfine headquarters (HQ) in Guilford, CT, USA. All healthy control imaging at Hyperfine HQ was performed under a protocol approved by the Western Institutional Review Board-Copernicus Group (WCG) local IRB and written informed consent was obtained prior to imaging. Each exam consisted of two-dimensional multi-slice, whole brain axial T2W and FLAIR sequence imaging. The authors affirm that all patient participants enrolled from July 2018 to March 2020 and non-patient human research participants enrolled at Hyperfine HQ provided informed consent for publication of the data in Table 1 and the images presented in Figs. 2 and 3. Patients receiving pMRI examinations as part of clinical care from March 2020 to November 2020 did not provide informed consent as the requirement for informed consent to publish the data was waived by the Yale Human Research Protection Program.
Portable MRI specifications. All patients and healthy controls were examined with a bedside pMRI system (Hyperfine Research, Inc., Guilford, CT, USA) at 0.064T static magnetic field strength. The pMRI device has a height of 140 cm, a width of 86 cm, and weighs 630 kg. The critical 5 Gauss (0.5 mT) boundary around the scanner extends into a circle with a diameter of 158 cm and is deployed when the device is in transit. The scanner is positioned at the head of the patient's hospital bed. The scanner bridge adjoins the hospital bed with the pMRI device and has a length of 35 cm. The vertical clearance between magnets and horizontal clearance between closed RF shields are 32 cm and 55 cm, respectively. The head coil has a length of 26 cm, a height of 26 cm, and a width of 20 cm (Fig. 1). The pMRI device can be transported by elevator and requires a doorway clearance >86 cm wide. Power was drawn from a standard 15A, 110 V wall outlet in the patient's hospital room or in the mock ICU at Hyperfine HQ. The decibel level inside the coil ranges from 60 to 80 db. The pMRI uses a biplanar, whole body (unshielded) gradient system which includes X, Y, and Z direction gradient coils, enabling imaging in any 3D orientation similar to conventional fixed-location MRI scanners. The peak amplitudes of the gradients are 25 mT/m on the X-and Y-axis and 26 mT/m on the Z-axis. The gradient peak slew rates are 23 T/m/s on the Xand Y-axis and 67 T/m/s on the Z-axis. Images were acquired using a single channel transmit, 8-channel receiver head coil.
The scanning environment contained ferrous metal and standard intensive care unit equipment, including but not limited to the electrocardiogram and vital signs monitor, IV infusion pumps, ventilators, compressed gas tanks, and dialysis machines. Exams were administered by research staff trained to operate the pMRI scanner without the need for a specialized MR technician. Scan sequences were controlled using a tablet computer interface (iPad Pro, 2nd generation and 3rd generation; Apple, Cupertino, CA, USA The scanner comes with a system quality assurance phantom, which is scanned on a monthly basis for calibration and monitoring purposes. The phantom image sets are uploaded to the Hyperfine Cloud Picture Archive and Communication System (PACS) and evaluated by Hyperfine Research, Inc. clinical scientists. In addition, each scanner passes a phantom-based factory acceptance test before delivery, which verifies performance metrics established according to National Electrical Manufacturers Association (NEMA) standards, and the results are reported to the U.S. FDA.
Sensitivity and specificity rater evaluation. A panel of two board-certified neuroradiologists (G.S. 40 years of experience, A.M. 22 years of experience) independently evaluated T2W and FLAIR pMRI exams. An ICH imaging core lab researcher (A.L.) with 7 years of experience reviewing clinical neuroimaging of ICH adjudicated the cases of disagreement between neuroradiologists.
Readers were blinded to all clinical and demographic information. The order of patients was randomized and both pulse sequences (T2W and FLAIR) were interpreted simultaneously in Horos (v3.3.5) or RadiAnt (v.2020.2.3). T2W and FLAIR imaging sequences were selected for rater evaluation since these were the only two sequences that were consistently available on the device across software changes occurring from the beginning of the study (July 2018) to its completion (November 2020).
For each pMRI exam, readers recorded the following data: presence of intracranial abnormality, whether the intracranial abnormality was an intraparenchymal lesion, presence of hemorrhage within the intraparenchymal lesion, supratentorial versus infratentorial location, side of intraparenchymal lesion, and presence of intraventricular hemorrhage (IVH). A positive score for ICH was determined by raters identifying an intracranial abnormality as an intraparenchymal lesion with hemorrhage in the correct location and anatomical side, in agreement with the available conventional radiologic report. The main outcome measure was the presence of ICH diagnosed on pMRI as determined by rater consensus. A.M.P., I.R.C.). Lesion annotation was completed manually by readers using Horos (v3.3.5) on T2W and FLAIR sequences. Readers annotated the hemorrhage core, hemorrhage rim (if distinguishable), and contralateral hemisphere on a single representative slice. Hemorrhage segmentations were used to compute signal intensity ratios (SIR).
Hematoma volume, localization, and patient outcome. Hematoma volume was measured by manual segmentation by one reader (M.H.M.) in AFNI (v21.1.02) and estimated using the ABC/2 method 44 by four neuroimaging research core lab readers (M.H.M., M.M.Y., A.M.P., I.R.C.) using Horos (v3.3.5). Raters annotated pMRI T2W and FLAIR sequences and the closet conventional imaging study (CT or 1.5/3T MRI T2W and FLAIR sequences) within 36 h of the pMRI exam. Lowfield pMRI exams whose closest conventional imaging study occurred outside 36 h were excluded from this sub-analysis to account for the highly dynamic nature of ICH 56,57 . For the ABC/2 method, lesions with non-ellipsoid morphologies were excluded due to the inaccuracy of the ABC/2 measurement in these cases 45  Statistical methods. Descriptive statistics are reported using means (SD) and medians (interquartile range [IQR]), as appropriate. Point estimates and 95% confidence intervals were determined for sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Interrater reliability for sensitivity and specificity was calculated for observations of both T2W and FLAIR exams using the Gwet's AC2 agreement coefficient 59,60 . SIR were computed by dividing the mean signal intensity of the annotated lesion (hemorrhage core and, if applicable, hemorrhage rim) on a single representative slice by the mean signal intensity of the contralateral hemisphere. Averaged SIR and standard deviations were computed across individual raters. SIR < 0.95 were characterized as hypointense, 0.95 ≤ SIR ≤ 1.05 were characterized as isointense, and SIR > 1.05 were characterized as hyperintense. For lesion localization across modalities, the Euclidean distance between the centroid of the segmented hemorrhage and each point ROI was calculated in RStudio v3.6.1 for each pMRI exam. The intraclass correlation coefficient (ICC) was used to assess the accuracy of pMRI hematoma volume measurements and Euclidean distances on T2W and FLAIR sequences compared to the closest conventional neuroimaging study measurements (CT or 1.5/3T MRI T2W and FLAIR). In addition, the ICC was used to assess interrater agreement between readers using the ABC/2 method to estimate hematoma volume. Manual and ABC/2 hematoma volumes on pMRI sequences (T2W and FLAIR) were correlated with National Institutes of Health Stroke Scale (NIHSS) obtained at the closest time to pMRI examination and Modified Rankin Scale (mRS) at discharge using unadjusted and adjusted linear regression models. Adjusted linear regression models included sex, race, and age. All statistical computations were completed using RStudio v3.6.1.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Source data are provided with this paper as a source data file. Raw data associated with Tables 1, 2, and 3 and Figs. 4 and 5 are included in the source data file. While a public portable MRI neuroimaging repository is not established yet, there are ongoing efforts to make the neuroimaging data publicly available. In the interim, the neuroimaging studies analyzed during the current study are available from the corresponding author upon request. Source data are provided with this paper.