Deep Learning-Generated Synthetic MR Imaging STIR Spine Images Are Superior in Image Quality and Diagnostically Equivalent to Conventional STIR: A Multicenter, Multireader Trial

BACKGROUND AND PURPOSE

Deep learning image reconstruction allows faster MR imaging acquisitions while matching or exceeding the standard of care and can create synthetic images from existing data sets. This multicenter, multireader spine study evaluated the performance of synthetically created STIR compared with acquired STIR.

MATERIALS AND METHODS

From a multicenter, multiscanner data base of 328 clinical cases, a nonreader neuroradiologist randomly selected 110 spine MR imaging studies in 93 patients (sagittal T1, T2, and STIR) and classified them into 5 categories of disease and healthy. A DICOM-based deep learning application generated a synthetically created STIR series from the sagittal T1 and T2 images. Five radiologists (3 neuroradiologists, 1 musculoskeletal radiologist, and 1 general radiologist) rated the STIR quality and classified disease pathology (study 1, n = 80). They then assessed the presence or absence of findings typically evaluated with STIR in patients with trauma (study 2, n = 30). The readers evaluated studies with either acquired STIR or synthetically created STIR in a blinded and randomized fashion with a 1-month washout period. The interchangeability of acquired STIR and synthetically created STIR was assessed using a noninferiority threshold of 10%.

RESULTS

For classification, there was a decrease in interreader agreement expected by randomly introducing synthetically created STIR of 3.23%. For trauma, there was an overall increase in interreader agreement by +1.9%. The lower bound of confidence for both exceeded the noninferiority threshold, indicating interchangeability of synthetically created STIR with acquired STIR. Both the Wilcoxon signed-rank and t tests showed higher image-quality scores for synthetically created STIR over acquired STIR (P < .0001).

CONCLUSIONS

Synthetically created STIR spine MR images were diagnostically interchangeable with acquired STIR, while providing significantly higher image quality, suggesting routine clinical practice potential.

Deep Learning Enables 60% Accelerated Volumetric Brain MRI While Preserving Quantitative Performance: A Prospective, Multicenter, Multireader Trial

BACKGROUND AND PURPOSE

In this prospective, multicenter, multireader study, we evaluated the impact on both image quality and quantitative image-analysis consistency of 60% accelerated volumetric MR imaging sequences processed with a commercially available, vendor-agnostic, DICOM-based, deep learning tool (SubtleMR) compared with that of standard of care.

MATERIALS AND METHODS

Forty subjects underwent brain MR imaging examinations on 6 scanners from 5 institutions. Standard of care and accelerated datasets were acquired for each subject, and the accelerated scans were enhanced with deep learning processing. Standard of care, accelerated scans, and accelerated-deep learning were subjected to NeuroQuant quantitative analysis and classified by a neuroradiologist into clinical disease categories. Concordance of standard of care and accelerated-deep learning biomarker measurements were assessed. Randomized, side-by-side, multiplanar datasets (360 series) were presented blinded to 2 neuroradiologists and rated for apparent SNR, image sharpness, artifacts, anatomic/lesion conspicuity, image contrast, and gray-white differentiation to evaluate image quality.

RESULTS

Accelerated-deep learning was statistically superior to standard of care for perceived quality across imaging features despite a 60% sequence scan-time reduction. Both accelerated-deep learning and standard of care were superior to accelerated scans for all features. There was no difference in quantitative volumetric biomarkers or clinical classification for standard of care and accelerated-deep learning datasets.

CONCLUSIONS

Deep learning reconstruction allows 60% sequence scan-time reduction while maintaining high volumetric quantification accuracy, consistent clinical classification, and what radiologists perceive as superior image quality compared with standard of care. This trial supports the reliability, efficiency, and utility of deep learning-based enhancement for quantitative imaging. Shorter scan times may heighten the use of volumetric quantitative MR imaging in routine clinical settings.

Differentiation of brain tumor-related edema based on 3D T1rho imaging

BACKGROUND AND PURPOSE

Cerebral edema associated with brain tumors is an important source of morbidity. Its type depends largely on the capillary ultra-structures of the histopathologic subtype of underlying brain tumor. The purpose of our study was to differentiate vasogenic edema associated with brain metastases and infiltrative edema related to diffuse gliomas using quantitative 3D T1 rho (T1ρ) imaging.

MATERIALS AND METHODS

Preoperative MR examination including whole brain 3D T1ρ imaging was performed in 23 patients with newly diagnosed brain tumors (9 with metastasis, 8 with lower grade glioma, LGG, 6 with glioblastoma, GBM). Mean T1ρ values were measured in regions of peritumoral non-enhancing T2 signal hyperintensity, excluding both enhancing and necrotic or cystic component, and normal-appearing white matter.

RESULTS

Mean T1ρ values were significantly elevated in the vasogenic edema surrounding intracranial metastases when compared to the infiltrative edema associated with either LGG or GBM (p=0.02 and <0.01, respectively). No significant difference was noted between T1ρ values of infiltrative edema between LGG and GBM (p=0.84 and 0.96, respectively).

CONCLUSION

Our study demonstrates the feasibility and potential diagnostic role of T1ρ in the quantitative differentiation between edema related to intracranial metastases and gliomas and as a potentially complementary tool to standard MR techniques in further characterizing pathophysiology of vasogenic and infiltrative edema.

Synthetic MRI for Clinical Neuroimaging: Results of the Magnetic Resonance Image Compilation (MAGiC) Prospective, Multicenter, Multireader Trial

BACKGROUND AND PURPOSE

Synthetic MR imaging enables reconstruction of various image contrasts from 1 scan, reducing scan times and potentially providing novel information. This study is the first large, prospective comparison of synthetic-versus-conventional MR imaging for routine neuroimaging.

MATERIALS AND METHODS

A prospective multireader, multicase noninferiority trial of 1526 images read by 7 blinded neuroradiologists was performed with prospectively acquired synthetic and conventional brain MR imaging case-control pairs from 109 subjects (mean, 53.0 ± 18.5 years of age; range, 19-89 years of age) with neuroimaging indications. Each case included conventional T1- and T2-weighted, T1 and T2 FLAIR, and STIR and/or proton density and synthetic reconstructions from multiple-dynamic multiple-echo imaging. Images were randomized and independently assessed for diagnostic quality, morphologic legibility, radiologic findings indicative of diagnosis, and artifacts.

RESULTS

Clinical MR imaging studies revealed 46 healthy and 63 pathologic cases. Overall diagnostic quality of synthetic MR images was noninferior to conventional imaging on a 5-level Likert scale (P < .001; mean synthetic-conventional, -0.335 ± 0.352; Δ = 0.5; lower limit of the 95% CI, -0.402). Legibility of synthetic and conventional morphology agreed in >95%, except in the posterior limb of the internal capsule for T1, T1 FLAIR, and proton-density views (all, >80%). Synthetic T2 FLAIR had more pronounced artifacts, including +24.1% of cases with flow artifacts and +17.6% cases with white noise artifacts.

CONCLUSIONS

Overall synthetic MR imaging quality was similar to that of conventional proton-density, STIR, and T1- and T2-weighted contrast views across neurologic conditions. While artifacts were more common in synthetic T2 FLAIR, these were readily recognizable and did not mimic pathology but could necessitate additional conventional T2 FLAIR to confirm the diagnosis.

Coupling between resting cerebral perfusion and EEG

While several studies have investigated interactions between the electroencephalography (EEG) and functional magnetic resonance imaging BOLD signal fluctuations, less is known about the associations between EEG oscillations and baseline brain haemodynamics, and few studies have examined the link between EEG power outside the alpha band and baseline perfusion. Here we compare whole-brain arterial spin labelling perfusion MRI and EEG in a group of healthy adults (n = 16, ten females, median age: 27 years, range 21-48) during an eyes closed rest condition. Correlations emerged between perfusion and global average EEG power in low (delta: 2-4 Hz and theta: 4-7 Hz), middle (alpha: 8-13 Hz), and high (beta: 13-30 Hz and gamma: 30-45 Hz) frequency bands in both cortical and sub-cortical regions. The correlations were predominately positive in middle and high-frequency bands, and negative in delta. In addition, central alpha frequency positively correlated with perfusion in a network of brain regions associated with the modulation of attention and preparedness for external input, and central theta frequency correlated negatively with a widespread network of cortical regions. These results indicate that the coupling between average EEG power/frequency and local cerebral blood flow varies in a frequency specific manner. Our results are consistent with longstanding concepts that decreasing EEG frequencies which in general map onto decreasing levels of activation.

P2-08-06: Improved Spatial Resolution Diffusion-Weighted Imaging for Characterizing Tumors and Treatment Response in Patients with Invasive Breast Cancer

Background: Diffusion weighted magnetic resonance imaging (DWI) is a non-invasive technique that is sensitive to tissue microstructure. Previous studies have shown that DWI adds positive predictive value in diagnostic studies of breast cancer and it has been shown to predict tumor response to neoadjuvant chemotherapy. While DWI shows promise for evaluating breast cancer, the technique suffers from limitations. Specifically, image distortion is common with the echo planar sequence available for DWI on clinical scanners, and spatial resolution is lower than that of other MRI sequences. Our group has optimized a high-resolution reduced field-of-view DWI acquisition, originally developed for the spine by Saritas et al., for breast imaging. The goal of this work was to compare high resolution (hr)-DWI) to standard resolution (std)-DWI for characterizing breast tumors.

Methods: Patients undergoing neoadjuvant chemotherapy were scanned with MRI before, during and after neoadjuvant chemotherapy as part of IRB-approved studies at our institution. Nine women were scanned with both hr-DWI and std-DWI before and after one cycle of chemotherapy. Apparent diffusion coefficient (ADC) maps were calculated from hr-DWI and std-DWI data using previously described methods. One tumor region of interest (ROI) was defined on the hr-DWI slice estimated to contain the largest tumor area. This tumor ROI was then applied to the corresponding slice and location on the std-DWI and hr-DWI ADC maps. Mean tumor ADC as well as 15th, 25th, 50th, 75th, and 90th percentile ADCs were calculated for both DWI acquisitions for all subjects.

Results: The mean tumor ADC values measured prior to treatment were similar for the hr-DWI and std-DWI acquisitions, however there was a significant difference between hr- and std-DWI 15th and 25th percentile ADC values (p= 0.0495, p=0.0717) For the early treatment time point, significant differences between the two DWI acquisitions were found for: mean tumor ADC, 15th, 25th, and 50th percentiles (p=0.0302, 0.0075, 0.0212, and 0.0488, respectively), with the most significant difference found for the lowest (15th) percentile measured. Tumor hr-DWI ADCs were consistently lower than std-DWI ADCs.

Discussion: These data show that although the mean ADC values calculated from the pre-treatment hr-DWI and std-DWI are similar, the lower percentile (15th, and 25th) ADC values are significantly lower for the hr-DWI acquisition. Our results also showed larger difference in lower percentile ADC values between the two sequences after one cycle of chemotherapy. The differences in the lower percentile ADC values calculated from the hr-DWI are consistent with reduced partial voluming between viable tumor tissue, which is characterized low ADC values, and normal fibroglandular tissue. This may be particularly important for post-treatment ADC measurements where tumor size may decrease, potentially making partial volume effects more pronounced. Continuing studies are evaluating the relationship between low percentile ADC values from hr-DWI and tumor stage and response to treatment.

Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P2-08-06.

Segmented k-space and real-time cardiac cine MR imaging with radial trajectories

The authors developed and evaluated two cine magnetic resonance (MR) imaging sequences with a radial rather than a rectilinear k-space coordinate frame: segmented k space and real-time true fast imaging with steady-state precession, or FISP. The two radial k-space segmentation (or view sharing) techniques, which were interleaved or continuous, were compared, and the feasibility of their application in cardiac cine MR imaging was explored in phantom and volunteer studies. Images obtained with the radial sequences were compared with those obtained with two-dimensional Fourier transform, or 2DFT, sequences currently used in cine MR imaging. Temporal resolution of 55 msec was achieved with the real-time radial sequences, which allowed acquisition of almost 19 high-quality images per second.

A method for fast 3D tracking using tuned fiducial markers and a limited projection reconstruction FISP (LPR-FISP) sequence

This work demonstrates the feasibility of using wireless, tuned fiducial markers with a limited projection reconstruction-fast imaging with steady-state free precession sequence (LPR-FISP) to accurately obtain tracking information necessary for interactive scan plane selection in magnetic resonance imaging (MRI). The position and orientation of a rigid interventional device can be uniquely determined from the 3D coordinates of three fiducial markers mounted in a known configuration on the device. Three fiducial markers were tuned to the proton resonant frequency in a 0.2T open MR scanner and mounted to the surface of a cylindrical water phantom. An LPR-FISP sequence was developed to suppress the water phantom signal while preserving that of the fiducial markers through a nonselective low-tip-angle excitation and a dephaser gradient applied prior to data acquisition. A localization algorithm was developed to accurately calculate the 3D coordinates of the fiducial markers using four LPR-FISP projections in two orthogonal scan planes. The sequence repetition time (TR = 21 msec) and the limited projection set resulted in fast LPR-FISP coordinate acquisition times of approximately 170 msec with an accuracy (max error) of 3 mm on a 0.2T MR system. This fast, accurate tracking method provides the fundamental technology for interactive MRI scan plane definition for rigid interventional devices without the need for stereotactic cameras or reference frames.

[Comparison of susceptibility artefacts of different radiofrequency electrodes at O.2 T. Influence of electrode positioning, pulse sequence and image reconstruction methods]

PURPOSE

Interventional MRI procedure monitoring requires small but accurate susceptibility artifacts of the instruments used. In this investigation, susceptibility artifacts of different RF-electrode designs were compared using a variety of pulse sequences and k-space acquisition methods.

METHODS

4 different 18-gauge RF-electrodes (with three single electrodes made of stainless steel, copper, inconal, and a triple-clustered electrode configuration made of inconal) were placed in a 0.2 T MR-scanner perpendicular to the main magnetic field. Pulse sequences used included: TSE T2, FISP, true-FISP, PSIF, and a temperature sensitive ES-GRE sequence. In addition to the 2D Cartesian k-space trajectory with Fourier transformation (2DFT), projection reconstruction (PR) was used with the FISP, true-FISP and PSIF sequences.

RESULTS

The best tip accuracy was achieved with the combination of inconal electrodes and TSE T2. The usefulness of the tested sequences was found to be: TSE T2 > PSIF > FISP/true-FISP > ES-GRE. In general 2DFT provided better or equal tip accuracy than PR. The apparent shaft width was smaller using the copper electrode compared to the inconal electrode. However, the "match shaped" tip artifact of the copper probe led to a higher error in tip accuracy.

CONCLUSIONS

TSE-T2 sequences and Cartesian 2DFT acquisitions should be used for accurate tip positioning at 0.2 T. Further, artifact size of the electrode shaft prevents the use of inconal for temperature sensitive sequences. Copper electrodes can be used for these purposes, although copper is not considered to be biocompatible at present.

Two-step navigatorless correction algorithm for radial k-space MRI acquisitions

A new way to correct magnetic resonance image artifacts resulting from view-dependent phase variations and view-dependent variations in rigid body object translation is presented by exploiting basic properties of the trajectory of radial k-space acquisitions. Simulations, phantom studies, and in vivo experiments are used to demonstrate the feasibility and the utility of this method. While somewhat analogous to navigator echo correction, in which special gradients are interleaved into the imaging sequence so echoes at the center of k-space can be acquired prior to or after collection of the image data, the current method does not require additional new gradient structures within the pulse sequence or increases in scan time. The new method uses the phase information from all collected radial k-space data points rather than only the navigator echo, which permits correction of multiple sources of view-dependent phase variation in the image data. The resultant effect is improved image quality in radial MRI acquisitions. Magn Reson Med 45:277-288, 2001.

Radial keyhole sequences for low field projection reconstruction interventional MRI

Interventional magnetic resonance imaging (IMRI) is a rapidly emerging application for MRI in which diagnostic and therapeutic procedures are performed with MR image guidance. Real-time or near-real-time image acquisition and relative insensitivity to motion are essential for most intraoperative, therapeutic, and diagnostic procedures performed under MR guidance. The purpose of this work was to demonstrate the development and utility of two alternative rapid acquisition strategies during IMRI that are analogous to computed tomography fluoroscopy or keyhole MRI in a radial rather than rectilinear coordinate frame. The two strategies discussed here, interleaved projection reconstruction and continuous projection reconstruction, are compared and the feasibility of their application in experimental interventional applications is studied. J. Magn. Reson. Imaging 2001;13:142-151.

Effects of superparamagnetic iron oxide on radio-frequency-induced temperature distribution: in vitro measurements in polyacrylamide phantoms and in vivo results in a rabbit liver model

PURPOSE

To determine whether contrast medium containing superparamagnetic iron oxide (SPIO) alters radio-frequency (RF)-related temperature distribution in acrylamide phantoms and in an in vivo model.

MATERIALS AND METHODS

In nine acrylamide phantoms with increasing SPIO content, RF was applied with simultaneous measurement of temperature profile along the probe track. Additionally, magnetic resonance imaging-guided RF ablation was performed in the liver of six rabbits after the intravenous administration of SPIO (0.05 mL per kilogram of body weight) 40 minutes prior to ablation (SPIO group) and in another six rabbits without prior SPIO administration (control group). Coagulation diameter was evaluated on the basis of postprocedural imaging and subsequent gross pathologic findings. Statistical analysis was performed with the Student t test.

RESULTS

In the phantoms, progressive increases in iron content resulted in higher temperatures along the RF electrode track (P < .05). In the in vivo model, however, SPIO at physiologic concentrations did not significantly increase the diameter of coagulation on the basis of either postprocedural imaging or subsequent gross pathologic findings. Additionally, no significant differences were seen in other RF-related parameters including impedance, voltage, current, and grounding pad temperature.

CONCLUSION

Administration of SPIO in conjunction with RF ablation of focal liver lesions is feasible and safe, but no significant difference in the extent of induced coagulation can be expected.

Temperature Measurement Using Echo-Shifted FLASH at Low Field for Interventional MRI

Temperature Measurement Using Echo-Shifted FLASH at Low Field for Interventional MRI. Yiu-Cho Chung, Jeffrey L. Duerk, Ajit Shankaranarayanan, Monika Hampke, Elmar M. Merkle, and Jonathan S. Lewin. (Article was originally published in the Journal of Magnetic Resonance Imaging, Volume 9, No. 1, 1999). In this article, some of the references were printed with the incorrect journal name. Here is the corrected list of references for this article.

Temperature measurement using echo-shifted FLASH at low field for interventional MRI

We investigated the feasibility of using echo-shifted fast low-angle shot (FLASH) for temperature-monitored thermo-therapeutic procedures in a 0.2 T interventional magnetic resonance (MR) scanner. Based on the proton resonance frequency shift technique, modified echo-shifted FLASH has sufficiently high signal-to-noise ratio to provide accurate temperature maps with short scan times, i.e., 5 seconds in phantoms (TR = 20.5 msec; effective TE = 30 msec; one echo shift; NSA = 2) and ex vivo experiments (TR = 19.4 msec; effective TE = 28.9 msec; one echo shift; NSA = 2) and 3 seconds (TR = 19.4 msec; effective TE = 28.9 msec, one echo shift; NSA 1) for an in vivo case. The proton resonance frequency shifts with temperature observed in a 0.2 T MR scanner using this sequence were -0.0072 ppm/degrees C (temperature uncertainty = +/-2.5 degrees C) for polyacrylamide phantoins and -0.0086 ppm/degrees C (temperature uncertainty = +/- 1 degrees C) for ex vivo bovine liver. These experiments demonstrated that echo-shifted FLASH is a viable method for low-field temperature monitoring despite the decreased signal and decreased phase sensitivity compared with its counterpart in a 1.5 T MR imaging system. The improved temporal resolution of temperature images, now possible in low-field interventional MR systems using echo-shifted FLASH, will allow clinicians more accurate monitoring of interstitial ablation in MR-guided interventional procedures.

Developing a multichannel temperature probe for interventional MRI

Interventional MRI (I-MRI) guided thermal tissue ablation has been used for a variety of interventional cancer therapies. These would be further facilitated by temperature-sensitive sequences on low magnetic field MR images. However, until these sequences have been reliably implemented at low fields, other methods of temperature measurement are required. This project describes the development of a low cost, reliable, MRI-compatible temperature sensor array useful at a temperature range from 37 degrees C to higher than 90 degrees C. The device uses a three-channel thermocouple sensor array connected to a variety of filtering and signal-conditioning electronics, analog-to-digital (A/D) converters, and personal computers. The sensors induce negligible field distortion. Similarly, no MRI-based measurement artifacts are observed. One-dimensional temperature profiles are generated with thermocouple signal linearization performed by the software.