Figure 1. Simplified schematic of decision-analytic model. The model was developed based on the clinical scenario where patients had undergone prior treatment for a brain tumor and now have new or persistent brain findings on imaging that are suspicious for recurrent tumor or cerebral necrosis⁵. It compared the methods of performing-MRS with not-performing-MRS where patient would go straight to biopsy.

Figure 2. Scatter plots of probabilistic sensitivity analysis for MRS vs No MRS for time horizon of 5yr, 10yr, and lifetime. Quadrant A: MRS performed is less effective and more expensive; Quadrant B: MRS performed is more effective and more expensive; Quadrant C: MRS performed is less effective and less expensive; Quadrant D: MRS performed is more effective and less expensive. The summary point (in orange) shows the average incremental cost and incremental QALY of that plot.

DWI at 0.55T. We compare [1,5,10,40] averages of (A) b=1000 and (B) ADC images for standard and RMT reconstructions. ADC images are scaled differently, as the Rician bias of standard images dramatically lowers ADC.

Coefficient of Variation (CoV) [std(ADC)/mean(ADC)] in ADC as a function of the number of averages compared between standard (green) and RMT(red). The smaller dots represent individual cases, whereas the larger dots represent the median CoV curve. RMT consistently shows lower CoV than the standard recon, even suggesting that variance due to noise can be minimized with this approach. For both reconstructions, CoV decreases with number of averages (except for cases with small ROIs [peripheral zone, lesion] due to motion between averages).

Figure 2: Mammograms with breast density estimation following the American College of Radiology Breast Imaging Reporting and Data System categories 1-4 with corresponding PDFF maps.

Figure 3: Correlation of mean PDFF values to the four American College of Radiology Breast Imaging Reporting and Data System categories with adjusted p-values revealing a significant inverse correlation.

Figure 1. Examples for standard-dose (SD-CT: Left), low-dose CT (LD-CT: Middle) and pulmonary MR imaging with UTE (UTE-MRI: Right) in Lung-RADS Category 2 (1st line), catefory 4A (2nd line) and Category 4X (3rd line).

Each method has no difference of radiological findings as well as Lung-RADS classification in each subject.

Figure 2. Results of JAFROC analysis for lung nodule detection.

There were no significant differences of figure of merit, true-positive ratio and false-positive ratio/case among all methods.

Figure 1: (a) Customized Ford Transit “Scan-a-Van” with (b) aluminum pallet for loading and unloading the scanner. (c) A customized roll cage was built into the van and welded to the frame to provide a docking point for the scanner pallet and to secure the scanner in place while driving. The roll cage also increases driver safety, preventing the scanner from entering the driver compartment in the case of sudden stopping. (d) A shielded battery was also developed to provide >6 hrs of battery power when needed.

Figure 2: Comparison and evaluation of image quality in the mobile scanner compared to a static device. We found no significant differences or biases in geometric distortion or image quality.

Images (a)-(d) and (e)-(f) demonstrate T2 FLAIR images processed with two different pipelines so as to independently explore their effects on radiologists’ confidence scores. Image (g) is the corresponding native DWI. The stroke pictured is that for which radiologists’ confidence averaged over all DWI-T2 FLAIR pairs was highest when answering questions related to acute stroke; average scores for presence of acute infarct, location of acute infarct, and DWI-T2 FLAIR mismatch were 5.0, 5.0, and 4.3, respectively.

Mean radiologist confidence score in identification of DWI-T2 FLAIR mismatch for acute ischemic stroke versus relative acquisition time. The data illustrate that reducing relative acquisition time by up to R=6 yields a decrease in radiologists' confidence in the accuracy of their identification of DWI-T2 FLAIR mismatch, at most, by 23%.

Figure 5. Example of motion-resolved reconstruction of mid-resolution images acquired at 0.55T using the free-running framework: The final result from the FRF reconstruction shows how, using the physiological signals extracted from the data, both cardiac and respiratory motion are well resolved. The last row shows the large anatomical coverage provided by the 3D radial acquisition.

Figure 1. Free Running Framework: (A) Data were continuously acquired using a prototype 3D radial trajectory regardless of cardiac or respiratory motion. (B) The principal components of the SI readouts were analyzed and filtered to extract cardiac and respiratory self-gating signals. (C) The motion signals were used to sort the data into motion-consistent cardiac and respiratory phases. (D) The binned data were then reconstructed with a k-t sparse SENSE algorithm that exploits data redundancy among the different bins followed by a high-dimensionality patch-based denoising.

Figure 1. Software and hardware stack of our OCRA package. A standard PulSeq description (.seq) file is converted by the ocra-pulseq and marcos_client libraries into data that can be executed on the open-source OCRA console. Yellow boxes show the novel elements in the stack: ocra-pulseq and marcos_client Python libraries, the new server and gradient firmware in the OCRA system, and the new OCRA1 DAC and GPA-FHDO gradient boards.

Figure 4. Example imaging data acquired with OCRA console. (a) 1D projection imaging of a 1-cm FOV two-tube phantom for use in MRI educational curriculum on a tabletop scanner, along with 2D FFT reconstruction. (b) K-space data and a 2D image of a star phantom are also shown. (c) Phase and magnitude of a 2D turbo spin echo acquisition of an apple, and (d) two slides of a 3D turbo spin echo acquisition of a forearm in vivo, both acquired on the Leiden brain scanner platform.

3D FLAIR images of a patient with a diagnosis of stroke, acquired at the bedside without RF shielded room. Images were reconstructed without (a) and with (b) the proposed interference rejection method. Average rejection was 29 dB (range 12-39 dB). Arrow points to pathology (distal M1 occlusion).

Example of portable scanner usage outside of a shielded room. Neurointensive Care Unit (a), with the following equipment surrounding the patient: IV drip, EKG, ancillary emergency equipment (vitals, oxygen/gas hook ups). Scanning a healthy volunteer outdoor (b) while operating the system with a power generator.

Figure 1: Left petrous Giant cell tumor. PseudoCT image (L) clearly depicting the lesion and its margins as compared with CT image (R).

Table 1: Results of the image analysis between PseudoCT and CT scan images.

Figure and Table comparing the temperature increase of the main magnet caused by running a bipolar trapezoidal gradient. The pulse sequence contained 40 cycles of 0.9ms duration bipolar lobes and had a maximum field strength of 71.4 mT/m in the x- and y- axis gradients and a 75.75 mT/m in the z-axis gradient.

Comparison of heating at different gradient strengths. The parameters of the fit are shown in Figure 2. The pulse sequence used in these trials was a single lobe Trapezoidal pulse with a rise time of 0.1ms, total duration of 200ms, and TR of 2000ms.

Figure 4. Sample examples of Spine labeling on 2D localizers for four different test subjects. DL predictions are shown as Red square for T12, green square for L4 and blue square for S1 respectively. A and B : Examples of cases with error much less than acceptable criteria ( < 8 mm error). C: Example with large L4 error and D: Example with large S1 error. Notice that error in one vertebra labelling in C and D doesn’t not affect the accuracy of other labels.

Figure 3. Centroid error for 50 test cases for T12, L4 and S1 labeling. Cutoff for acceptable performance ( 8 mm ) is shown in dotted lines. Most of the errors are within the threshold limits. Overall, accuracy for T12, L4 and S1 is 92%, 98% and 96% respectively. Arrows indicate cases C and D shown in Figure 4 for L4 and S1 error respectively.

Figure-1: Summary of the stroke protocol under use

Figure-2: MRI findings summarized

Figure 1. Workflow for converting a 2D Gradient Recalled Echo (GRE) sequence. A 2D GRE sequence designed in Prospa (G_pr) was first manually programmed in PyPulseq (G_py). Next, this PyPulseq 2D GRE was converted to a Prospa-friendly format (G_s2p) using seq2prospa. Finally, the two 2D GRE sequences (G_pr, G_s2p) were executed on a Kea² spectrometer to image a human hand and a structural phantom. The acquired images (I_pr, I_s2p) were reconstructed and visualized.

Figure 3. Image reconstructions of imaging the structural phantom using the standard Prospa and PyPulseq-converted sequences. The SNR values measured from the standard Prospa acquisition (a) and the PyPulseq-converted acquisition (b) were 32 dB and 34 dB respectively.

Fig 2. Canonical scores calculated with training cohort place testing cohort (Prior-to-Dx) between Healthy and Time-of-Dx.

Fig 5. Testing and validation combination. a). Overall patient survival status with the M+SE threshold (dashed line) calculated using all the Prior-to-Dx cases from both cohorts. b). The Kaplan-Meier survival predictions for all stage I and IIA cases. c). Significant Kaplan-Meier survival rates predicted using this threshold according to the date of patients Time-of-Dx, with detailed statistical parameters listed.

Outline of stage-dependent incidence of CC, early recurrence, and potential MRI-based treatment changes. Assuming that early recurrence represents disease that was occult on staging imaging, the incidence of MRI-dependent treatment changes can be estimated. For example, Stage III patients have a reported incidence of early recurrence between 2-15%. Assuming an MRI sensitivity of 95%, MRI liver staging at diagnosis may have detected occult disease in 1 in 52 patients (assuming 2% early recurrence) or 1 in 7 patients (assuming 15% early recurrence).

Incidence of CC by Stage and Incidence of Early and Late Recurrence: First column: incidence of CC by stage. Second column: incidence of early recurrence (recurrence < 12 months from curative intent surgery), by stage. For example, 50% of patients with resectable stage IV disease at diagnosis will have early recurrence. Third column: incidence of recurrence at 5 years. For example, 60-80% of patients with resectable stage IV disease at diagnosis will recur within 5 years.

Figure 1. SPGR and MP-RAGE images acquired at three different spiral trajectory lengths (measured by ADC time) show equivalent T1 contrast and SNR, when the number of spiral arms adjusted to match SNR and the flip angle was adjusted to keep TA constant. The T1 contrast was measured as the ratio of WM and GM signals (WM/GM), shown within the figure. The gray and white matter SNR was quantitatively measured and shown in Figure 2.

Figure 2. SPGR and MP-RAGE scans show equivalent gray matter and white matter SNR across scans having different ADC time when the number of spiral arms was adjusted to match the SNR. In comparison, the SNR drops in the SNR-unmatched reference scans, which were acquired without such adjustment. (Shown for the two subjects for SPGR and one subject for MP-RAGE)

Figure 2: Open-source qualitative IRSE and TSE sequences are compared to vendor gold standards. Parameters are FOV = 250 mm, N = 256, slice thickness = 5 mm, and FA = 90 deg. IRSE has TR = 2000 ms, TE = 12 ms, TI = 100 ms (125 ms for Siemens); TSE has TR = 2741 ms, TE = 50 ms, turbo factor = 4

Figure 3: T₁ and T₂ maps obtained from IRSE and TSE sequences (N = 128, FOV = 250 mm, slice thickness = 6 mm, FA = 90, 180 deg). The Pearson coefficient of correlation is 0.9557 for T₁ spheres (0.9983 for spheres 1-13) and -0.0849 for T₂ spheres (0.9994 for spheres 1-10). Shorter TIs and TEs than applied are required to accurately map the highest-index spheres.

Figure 1. What MRI phantoms/test objects does your institution/centre have available?

Figure 3: OSIPI’s mission to support open perfusion science is done through the 6 aims as outlined here each with the associated task forces.

Figure 1: A word cloud of the comments (produced by wordart.com). Words such as “standard”, “evaluation”, “tested”, “validated” indicate a consensus that instead of another software platform, the community really needs perfusion to be benchmarked and standardized.

Figure 1. MRI studies of the lumbar spine obtained at 0.55T with a maximum gradient strength of 25 mT/m and slew rate of 40 T/m/s2, 0.55T with a maximum gradient strength of 45 mT/m and slew rate of 200 T/m/s2, and 1.5T with a maximum gradient strength of 45 mT/m and slew rate of 200 T/m/s2. MR images demonstrate an L4-L5 disc protrusion (arrows), causing thecal sac deformity and left lateral recess narrowing

Figure 2. MRI studies of the lumbar spine obtained at 0.55T with a maximum gradient strength of 25 mT/m and slew rate of 40 T/m/s2, 0.55T with a maximum gradient strength of 45 mT/m and slew rate of 200 T/m/s2, and 1.5T with a maximum gradient strength of 45 mT/m and slew rate of 200 T/m/s2. MR images demonstrate congenital lumbar spinal stenosis with epidural lipomatosis causing thecal sac narrowing (arrows) to various degrees.

Figure 4: FLAIR (a-d) and T2-W (e-h) imaging of a recovering patient with a right MCA syndrome with multiple infarcts. Four time points were imaged over the course of 19 days, where the far left column is time point 1, and the far right column shows time point 4. Stable T2-W/FLAIR hyperintensities (yellow arrow), indicating mild encephalomalacia, were observed at all time points. Right ventricular shunt catheter is present (blue arrow).

Figure 1: Portable, point-of-care MRI deployed in a rehabilitation center, as demonstrated through (a) a photograph of the device docked behind the patient’s bed and (b) a zoomed-in photograph of the device.

Figure 2. ISD workflow: Using the left panel the ExploreASL plugin (green frame) can be selected, which then prompts the user to select the T1, FLAIR and ASL series from the patient’s image list (blue frame). In the bottom part of the panel the analysis quality is defined and key ASL acquisition parameters are set (red frame), allowing the ASL processing to match the used scan parameters.

Figure 3. Visual output from the ISD plugin. The quantified CBF map (right) is co-registered with 3D T1 (middle), allowing for correct localization and visualization of T1-based Region Of Interest (ROI) on the CBF map. The left panel shows the quality report for the processing pipeline that allows for quick visual check of segmentation and provides CBF values for GM and WM.

Figure 1: Photograph of the completed amplifier, including current monitors and power supply voltage readouts on the front panel.

Figure 5: In-vivo image of the lower leg acquired on a 50 mT MRI scanner using the described gradient amplifier to drive the gradient coils. The image is reconstructed from a 3D TSE acquisition and has a resolution of 1x1x4mm.

Figure 1. Example of two T2-weighted images of same prostate taken ten months apart

Figure 4. DCE-MRI, DWI, and PK %RC values

Figure 5: An example of IQT enhancement on real LMIC patient data for T1w, T2w and FLAIR images. Middle cerebral artery ischemia was detected involving the perirolanding regions. The lesion area highlighted by the red arrows shows hyperintense signal in T2w and FLAIR, and hypointense signal in T1w.

Figure 4: An example of IQT enhancement on real LMIC T1w, T2w and FLAIR images from a normal brain subject. The 1.5T reference has a different orientation with 0.36T input and IQT enhancement. The demonstrated three sequences were enhanced to 3T-like image quality with isotropic voxels. Low-field images show cross-talk and field inhomogeneity artifacts, but they were corrected during IQT Pre-processing.

Figure 1: Routine clinical brain protocol using conventional reconstruction method and resolution (top row) and DLR method with higher resolution (bottom row). The contrast changed between the DLR and CR was minimal, indicating that the DLR is a reliable technique to achieve higher image quality in shorter scan while maintaining the clinical image contrast.

Figure 2: Routine clinical knee protocol using conventional reconstruction method and resolution (top row) and DLR method with higher resolution (bottom row).

Figure 1. Demonstration of Deep Learning Reconstruction Technique

A. Parallel imaging acceleration factor of 4, conventional Fourier transform reconstruction, 78% reduction in acquisition time from the baseline sequence with increased noise.

B. Same acquisition as in figure A, deep learning reconstruction demonstrating reduction of apparent noise to less than baseline exam with DL reconstruction technique

C. Magnified detail view of A.

D. Magnified detail view of B.

Figure 2. T2 Weighted Comparisons

A. Baseline T2w FSE – time to acquire 2:24

B. R=2 T2w FSE – time to acquire 1:20

C. R=3 T2w FSE – time to acquire 1:20

D. R=4 T2w FSE – time to acquire 0:48

E. NEX=1 T2w FSE – time to acquire 1:20

F: Reduced Matrix T2w FSE – time to acquire 1:52

Figure 3. Comparison of gradient times during the pre-implementation period versus post- implementation period.

Figure 1. Gradient times of conventional versus fast brain MRI sequences.

Figure 2: Comparison of overall image quality for T2WI, DWI, and T1WI on low-field regular gradients, low-field lower adjusted gradients, and 1.5T scanner.

Figure 3: Representative single axial T2WI in the same volunteer imaged on 1.5T, LF-regular, and LF-lower adjusted gradients.