Improving Generalizability of PET DL Algorithms: List-Mode Reconstructions Improve DOTATATE PET Hepatic Lesion Detection Performance.

Deep learning (DL) algorithms used for DOTATATE PET lesion detection typically require large, well-annotated training datasets. These are difficult to obtain due to low incidence of gastroenteropancreatic neuroendocrine tumors (GEP-NETs) and the high cost of manual annotation. Furthermore, networks trained and tested with data acquired from site specific PET/CT instrumentation, acquisition and processing protocols have reduced performance when tested with offsite data. This lack of generalizability requires even larger, more diverse training datasets. The objective of this study is to investigate the feasibility of improving DL algorithm performance by better matching the background noise in training datasets to higher noise, out-of-domain testing datasets. 68Ga-DOTATATE PET/CT datasets were obtained from two scanners: Scanner1, a state-of-the-art digital PET/CT (GE DMI PET/CT; n = 83 subjects), and Scanner2, an older-generation analog PET/CT (GE STE; n = 123 subjects). Set1, the data set from Scanner1, was reconstructed with standard clinical parameters (5 min; Q.Clear) and list-mode reconstructions (VPFXS 2, 3, 4, and 5-min). Set2, data from Scanner2 representing out-of-domain clinical scans, used standard iterative reconstruction (5 min; OSEM). A deep neural network was trained with each dataset: Network1 for Scanner1 and Network2 for Scanner2. DL performance (Network1) was tested with out-of-domain test data (Set2). To evaluate the effect of training sample size, we tested DL model performance using a fraction (25%, 50% and 75%) of Set1 for training. Scanner1, list-mode 2-min reconstructed data demonstrated the most similar noise level compared that of Set2, resulting in the best performance (F1 = 0.713). This was not significantly different compared to the highest performance, upper-bound limit using in-domain training for Network2 (F1 = 0.755; p-value = 0.103). Regarding sample size, the F1 score significantly increased from 25% training data (F1 = 0.478) to 100% training data (F1 = 0.713; p < 0.001). List-mode data from modern PET scanners can be reconstructed to better match the noise properties of older scanners. Using existing data and their associated annotations dramatically reduces the cost and effort in generating these datasets and significantly improves the performance of existing DL algorithms. List-mode reconstructions can provide an efficient, low-cost method to improve DL algorithm generalizability.

Learning Without Real Data Annotations to Detect Hepatic Lesions in PET Images.

OBJECTIVE: Deep neural networks have been recently applied to lesion identification in fluorodeoxyglucose (FDG) positron emission tomography (PET) images, but they typically rely on a large amount of well-annotated data for model training. This is extremely difficult to achieve for neuroendocrine tumors (NETs), because of low incidence of NETs and expensive lesion annotation in PET images. The objective of this study is to design a novel, adaptable deep learning method, which uses no real lesion annotations but instead low-cost, list mode-simulated data, for hepatic lesion detection in real-world clinical NET PET images. METHODS: We first propose a region-guided generative adversarial network (RG-GAN) for lesion-preserved image-to-image translation. Then, we design a specific data augmentation module for our list-mode simulated data and incorporate this module into the RG-GAN to improve model training. Finally, we combine the RG-GAN, the data augmentation module and a lesion detection neural network into a unified framework for joint-task learning to adaptatively identify lesions in real-world PET data. RESULTS: The proposed method outperforms recent state-of-the-art lesion detection methods in real clinical 68Ga-DOTATATE PET images, and produces very competitive performance with the target model that is trained with real lesion annotations. CONCLUSION: With RG-GAN modeling and specific data augmentation, we can obtain good lesion detection performance without using any real data annotations. SIGNIFICANCE: This study introduces an adaptable deep learning method for hepatic lesion identification in NETs, which can significantly reduce human effort for data annotation and improve model generalizability for lesion detection with PET imaging.

Modeling contrast-to-noise ratio from list mode reconstructions of 68Ga DOTATATE PET/CT: predicting detectability of hepatic metastases in shorter acquisition PET reconstructions.

BACKGROUND: Deep learning (DL) algorithms have shown promise in identifying and quantifying lesions in PET/CT. However, the accuracy and generalizability of these algorithms relies on large, diverse datasets which are time and labor intensive to curate. Modern PET/CT scanners may acquire data in list mode, allowing for multiple reconstructions of the same datasets with different parameters and imaging times. These reconstructions may provide a wide range of image characteristics to increase the size and diversity of datasets. Training algorithms with shorter imaging times and higher noise properties requires that lesions remain detectable. The purpose of this study is to model and predict the contrast-to-noise ratio (CNR) for shorter imaging times based on CNR from longer duration, lower noise images for 68Ga DOTATATE PET hepatic lesions and identify a threshold above which lesions remain detectable. METHODS: 68Ga DOTATATE subjects (n=20) with hepatic lesions were divided into two subgroups. The "Model" group (n=4 subjects; n=9 lesions; n=36 datapoints) was used to identify the relationship between CNR and imaging time. The "Test" group (n=16 subjects; n=44 lesions; n=176 datapoints) was used to evaluate the prediction provided by the model. RESULTS: CNR plotted as a function of imaging time for a subset of identified subjects was very well fit with a quadratic model. For the remaining subjects, the measured CNR showed a very high linear correlation with the predicted CNR for these lesions (R2 > 0.97) for all imaging durations. From the model, a threshold CNR=6.9 at 5-minutes predicted CNR > 5 at 2-minutes. Visual inspection of lesions in 2-minute images with CNR above the threshold in 5-minute images were assessed and rated as a 4 or 5 (probably positive or definitely positive) confirming 100% lesion detectability on the shorter 2-minute PET images. CONCLUSIONS: CNR for shorter DOTATATE PET imaging times may be accurately predicted using list mode reconstructions of longer acquisitions. A threshold CNR may be applied to longer duration images to ensure lesion detectability of shorter duration reconstructions. This method can aid in the selection of lesions to include in novel data augmentation techniques for deep learning.

Volumetric abdominal perfusion measurement using a pseudo-randomly sampled 3D fast-spin-echo (FSE) arterial spin labeling (ASL) sequence and compressed sensing reconstruction.

PURPOSE: To improve image quality and spatial coverage for abdominal perfusion imaging by implementing an arterial spin labeling (ASL) sequence that combines variable-density 3D fast-spin-echo (FSE) with Cartesian trajectory and compressed-sensing (CS) reconstruction. METHODS: A volumetric FSE sequence was modified to include background-suppressed pseudo-continuous ASL labeling and to support variable-density (VD) Poisson-disk sampling for acceleration. We additionally explored the benefits of center oversampling and variable outer k-space sampling. Fourteen healthy volunteers were scanned on a 3T scanner to test acceleration factors as well as the various sampling schemes described under synchronized-breathing to limit motion issues. A CS reconstruction was implemented using the BART toolbox to reconstruct perfusion-weighted ASL volumes, assessing the impact of acceleration, different reconstruction, and sampling strategies on image quality. RESULTS: CS acceleration is feasible with ASL, and a strong renal perfusion signal could be observed even at very high acceleration rates (≈15). We have shown that ASL k-space complex subtraction was desirable before CS reconstruction. Although averaging of multiple highly accelerated images helped to reduce artifacts from physiologic fluctuations, superior image quality was achieved by interleaving of different highly undersampled pseudo-random spatial sampling patterns and using 4D-CS reconstruction. Combination of these enhancements produces high-quality ASL volumes in under 5 min. CONCLUSIONS: High-quality isotropic ASL abdominal perfusion volumes can be obtained in healthy volunteers with a VD-FSE and CS reconstruction. This lays the groundwork for future developments toward whole abdomen free-breathing non-contrast perfusion imaging.

2D multi-spectral imaging for fast MRI near metal.

PURPOSE: To develop a fast 2D method for MRI near metal with reduced B0 in-plane and through-slice artifacts. METHODS: Multi-spectral imaging (MSI) approaches reduce artifacts in MR images near metal, but require 3D imaging of multiple excited volumes regardless of imaging geometry or artifact severity. The proposed 2D MSI method rapidly excites a limited slice and spectral region using gradient reversal between excitation and refocusing pulses, then uses standard 2D imaging, with the process repeating to cover multiple spectral offsets that are combined as in other MSI techniques. 2D MSI was implemented in a spin-echo-train sequence and validated in phantoms and in vivo by comparing it with standard spin-echo imaging and existing MSI techniques. RESULTS: 2D MSI images for each spatial-spectral region follow isocontours of the dipole-like B0 field variation, and thus frequency variation, near metal devices. Artifact correction in phantoms and human subjects with metal is comparable to 3D MSI methods, and superior to standard spin-echo techniques. Scan times are reduced compared with 3D MSI methods in cases where a limited number of slices are needed, though signal-to-noise ratio is also reduced as expected. CONCLUSION: 2D MSI offers a fast and flexible alternative to 3D MSI for artifact reduction near metal. Magn Reson Med 79:968-973, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Increased Speed and Image Quality for Pelvic Single-Shot Fast Spin-Echo Imaging with Variable Refocusing Flip Angles and Full-Fourier Acquisition.

Purpose To assess image quality and speed improvements for single-shot fast spin-echo (SSFSE) with variable refocusing flip angles and full-Fourier acquisition (vrfSSFSE) pelvic imaging via a prospective trial performed in the context of uterine leiomyoma evaluation. Materials and Methods Institutional review board approval and informed consent were obtained. vrfSSFSE and conventional SSFSE sagittal and coronal oblique acquisitions were performed in 54 consecutive female patients referred for 3-T magnetic resonance (MR) evaluation of known or suspected uterine leiomyomas. Two radiologists who were blinded to the image acquisition technique semiquantitatively scored images on a scale from -2 to 2 for noise, image contrast, sharpness, artifacts, and perceived ability to evaluate uterine, ovarian, and musculoskeletal structures. The null hypothesis of no significant difference between pulse sequences was assessed with a Wilcoxon signed rank test by using a Holm-Bonferroni correction for multiple comparisons. Results Because of reductions in specific absorption rate, vrfSSFSE imaging demonstrated significantly increased speed (more than twofold, P < .0001), with mean repetition times compared with conventional SSFSE imaging decreasing from 1358 to 613 msec for sagittal acquisitions and from 1494 to 621 msec for coronal oblique acquisitions. Almost all assessed image quality and perceived diagnostic capability parameters were significantly improved with vrfSSFSE imaging. These improvements included noise, sharpness, and ability to evaluate the junctional zone, myometrium, and musculoskeletal structures for both sagittal acquisitions (mean values of 0.56, 0.63, 0.42, 0.56, and 0.80, respectively; all P values < .0001) and coronal oblique acquisitions (mean values of 0.81, 1.09, 0.65, 0.93, and 1.12, respectively; all P values < .0001). For evaluation of artifacts, there was an insufficient number of cases with differences to allow statistical testing. Conclusion Compared with conventional SSFSE acquisition, vrfSSFSE acquisition increases 3-T imaging speed via reduced specific absorption rate and leads to significant improvements in perceived image quality and perceived diagnostic capability when evaluating pelvic structures. © RSNA, 2016 Online supplemental material is available for this article.

Image-based gradient non-linearity characterization to determine higher-order spherical harmonic coefficients for improved spatial position accuracy in magnetic resonance imaging.

PURPOSE: Spatial position accuracy in magnetic resonance imaging (MRI) is an important concern for a variety of applications, including radiation therapy planning, surgical planning, and longitudinal studies of morphologic changes to study neurodegenerative diseases. Spatial accuracy is strongly influenced by gradient linearity. This work presents a method for characterizing the gradient non-linearity fields on a per-system basis, and using this information to provide improved and higher-order (9th vs. 5th) spherical harmonic coefficients for better spatial accuracy in MRI. METHODS: A large fiducial phantom containing 5229 water-filled spheres in a grid pattern is scanned with the MR system, and the positions all the fiducials are measured and compared to the corresponding ground truth fiducial positions as reported from a computed tomography (CT) scan of the object. Systematic errors from off-resonance (i.e., B0) effects are minimized with the use of increased receiver bandwidth (±125kHz) and two acquisitions with reversed readout gradient polarity. The spherical harmonic coefficients are estimated using an iterative process, and can be subsequently used to correct for gradient non-linearity. Test-retest stability was assessed with five repeated measurements on a single scanner, and cross-scanner variation on four different, identically-configured 3T wide-bore systems. RESULTS: A decrease in the root-mean-square error (RMSE) over a 50cm diameter spherical volume from 1.80mm to 0.77mm is reported here in the case of replacing the vendor's standard 5th order spherical harmonic coefficients with custom fitted 9th order coefficients, and from 1.5mm to 1mm by extending custom fitted 5th order correction to the 9th order. Minimum RMSE varied between scanners, but was stable with repeated measurements in the same scanner. CONCLUSIONS: The results suggest that the proposed methods may be used on a per-system basis to more accurately calibrate MR gradient non-linearity coefficients when compared to vendor standard corrections.

Increased speed and image quality in single-shot fast spin echo imaging via variable refocusing flip angles.

PURPOSE: To develop and validate clinically a single-shot fast spin echo (SSFSE) sequence utilizing variable flip angle refocusing pulses to shorten acquisition times via reductions in specific absorption rate (SAR) and improve image quality. MATERIALS AND METHODS: A variable refocusing flip angle SSFSE sequence (vrfSSFSE) was designed and implemented, with simulations and volunteer scans performed to determine suitable flip angle modulation parameters. With Institutional Review Board (IRB) approval/informed consent, patients referred for 3T abdominal magnetic resonance imaging (MRI) were scanned with conventional SSFSE and either half-Fourier (n = 25) or full-Fourier vrfSSFSE (n = 50). Two blinded radiologists semiquantitatively scored images on a scale from -2 to 2 for contrast, noise, sharpness, artifacts, cardiac motion-related signal loss, and the ability to evaluate the pancreas and kidneys. RESULTS: vrfSSFSE demonstrated significantly increased speed (∼2-fold, P < 0.0001). Significant improvements in image quality parameters with full-Fourier vrfSSFSE included increased contrast, sharpness, and visualization of pancreatic and renal structures with higher bandwidth technique (mean scores 0.37, 0.83, 0.62, and 0.31, respectively, P ≤ 0.001), and decreased image noise and improved visualization of renal structures when used with an equal bandwidth technique (mean scores 0.96 and 0.35, respectively, P < 0.001). Increased cardiac motion-related signal loss with full-Fourier vrfSSFSE was seen in the pancreas but not the kidney. CONCLUSION: vrfSSFSE increases speed at 3T over conventional SSFSE via reduced SAR, and when combined with full-Fourier acquisition can improve image quality, although with some increased sensitivity to cardiac motion-related signal loss.

Faster pediatric 3-T abdominal magnetic resonance imaging: comparison between conventional and variable refocusing flip-angle single-shot fast spin-echo sequences.

BACKGROUND: Single-shot fast spin echo (SSFSE) is particularly appealing in pediatric patients because of its motion robustness. However radiofrequency energy deposition at 3 tesla forces long pauses between slices, leading to longer scans, longer breath-holds and more between-slice motion. OBJECTIVE: We sought to learn whether modulation of the SSFSE refocusing flip-angle train could reduce radiofrequency energy deposition without degrading image quality, thereby reducing inter-slice pauses and overall scan times. MATERIALS AND METHODS: We modulated the refocusing flip-angle train for SSFSE to minimize energy deposition while minimizing blurring and motion-related signal loss. In a cohort of 50 consecutive patients (25 boys, mean age 5.5 years, range 1 month to 17 years) referred for abdominal MRI we obtained standard SSFSE and variable refocusing flip-angle (vrfSSFSE) images and recorded sequence scan times. Two readers independently scored the images in blinded, randomized order for noise, tissue contrast, sharpness, artifacts and left lobe hepatic signal uniformity on a four-point scale. The null hypothesis of no difference between SSFSE and vrfSSFSE image-quality was assessed with a Mann-Whitney U test, and the null hypothesis of no scan time difference was assessed with the paired t-test. RESULTS: SSFSE and vrfSSFSE mean acquisition times were 54.3 and 26.2 s, respectively (P-value <0.0001). For each reader, SSFSE and vrfSSFSE noise, tissue contrast, sharpness and artifacts were not significantly different (P-values 0.18-0.86). However, SSFSE had better left lobe hepatic signal uniformity (P < 0.01, both readers). CONCLUSION: vrfSSFSE is twice as fast as SSFSE, with equivalent image quality with the exception of left hepatic lobe signal heterogeneity.

MRI accurately detects chronic exertional compartment syndrome: a validation study.

OBJECTIVE: To validate the diagnostic performance of an in-scanner exercise-based magnetic resonance imaging (MRI) examination used to screen for chronic exertional compartment syndrome (CECS). Final clinical impression and intracompartmental needle manometry (INM) served as the reference standards. MATERIALS AND METHODS: Consecutive patients, referred by a sports medicine physician or orthopedic surgeon, underwent the MRI examination for lower extremity pain over the past 4 years and 9 months. Utilizing a diagnostic T2-weighted intensity ratio threshold of 1.54, established by a prior cohort of patients, sensitivity, specificity, predictive value, and diagnostic odds ratio were calculated for the anterior compartments. The means of the T2-weighted intensity ratios were compared using the Wilcoxon rank sum test. RESULTS: A total of 79 patients were identified, and 76 met the inclusion criteria and were evaluated. Of these, 23 met clinical diagnostic criteria. Sensitivity and specificity were 96% (95% CI: 79-99%) and 87% (95% CI: 75-94%) using the established threshold of 1.54. T2-weighted intensity ratio provided excellent discrimination with a concordance statistic of 0.96 (95% CI: 0.91-1.00). In the subset of 36 patients with INM results, 23 patients met criteria for CECS, although only 19 patients met both INM and clinical criteria. The sensitivity and specificity of the MRI examination relative to INM results were 87% (95%: 70-96%) and 62% (95% CI: 36-82%) respectively. CONCLUSION: In-scanner exercise-based MRI demonstrated reliability and reproducibility as a non-invasive screening test for CECS, thus reducing the need for invasive INM.

Magnetic Resonance Elastography.

Often compared to the practice of manual palpation, magnetic resonance elastography is an emerging technology for quantitatively assessing the mechanical properties of tissue as a basis for characterizing disease. The potential of MRE as a diagnostic tool is rooted in the fact that normal and diseased tissues often differ significantly in terms of their intrinsic mechanical properties. MRE uses magnetic resonance imaging (MRI) in conjunction with the application of mechanical shear waves to probe tissue mechanics. This process can be broken down into three essential steps: inducing shear waves in the tissue,imaging the propagating shear waves with MRI, andanalyzing the wave data to generate quantitative images of tissue stiffness MRE has emerged as a safe, reliable and noninvasive method for staging hepatic liver fibrosis, and is now used in some locations as an alternative to biopsy. MRE is also being used in the ongoing investigations of numerous other organs and tissues, including, for example, the spleen, kidney, pancreas, brain, heart, breast, skeletal muscle, prostate, vasculature, lung, spinal cord, eye, bone, and cartilage. In the article that follows, some fundamental techniques and applications of MRE are summarized.

MR elastography of the ex vivo bovine globe.

PURPOSE: To evaluate the feasibility of using MR elastography (MRE) to assess the mechanical properties of the eye. MATERIALS AND METHODS: The elastic properties of the corneoscleral shell of an intact, enucleated bovine globe specimen were estimated using MRE and finite element modeling (FEM), assuming linear, isotropic behavior. The two-dimensional (2D), axisymetric model geometry was derived from a segmented 2D MR image, and estimations of the Young's modulus in both the cornea and sclera were made at various intraocular pressures using an iterative flexural wave speed matching algorithm. RESULTS: Estimated values of the Young's moduli of the cornea and sclera varied from 40 to 185 kPa and 1 to 7 MPa, respectively, over an intraocular pressure range of 0.85 to 9.05 mmHg (1.2 to 12.3 cmH(2)O). They also varied exponentially as functions of both wave speed and intraocular dP/dV, an empirical measure of "ocular rigidity." CONCLUSION: These results show that it is possible to estimate the intrinsic elastic properties of the corneoscleral shell in an ex vivo bovine globe, suggesting that MRE may provide a useful means to assess the mechanical properties of the eye and its anatomy. Further development of the technique and modeling process will enhance its potential, and further investigations are needed to determine its clinical potential.

Chronic exertional compartment syndrome of the lower extremities: improved screening using a novel dual birdcage coil and in-scanner exercise protocol.

OBJECTIVE: The purpose of this study was to design and evaluate an MRI screening protocol for chronic exertional compartment syndrome (CECS) of the lower legs using an in-scanner exercise protocol and novel dual birdcage coil design for improved imaging. MATERIALS AND METHODS: Coil and phantom studies: a custom-made dual birdcage coil designed for this protocol was evaluated for uniformity and signal-to-noise ratio (SNR) compared with a conventional phased-array receive-only torso coil and the body coil. Phantom and normal subject studies were performed to confirm coil performance. In-vivo studies: eight unaffected subjects and 42 patients with lower extremity symptoms suggestive of CECS were imaged with the dual birdcage coil and an in-scanner exercise protocol which included imaging at rest, during isometric resisted dorsi flexion, at rest (recovery), during isometric resisted plantar flexion and, again, at rest. Of 42 patients, 14 had confirmed CECS and 28 had lower extremity anomalies attributable to other causes. Ratios of relative T2-weighted signal intensities were calculated for exercise and recovery images compared to baseline after processing of images, including re-registration for motion, smoothing and segmentation to remove bone and pulsation artifacts from blood vessels. RESULTS: Receiver operating characteristic (ROC) analysis showed a threshold for the ratio of relative T2-weighted signal intensity of 1.54 to have a sensitivity of 96%, specificity of 90% and accuracy of 96% for CECS. Patients with CECS had their peak ratio of signal intensity compared with baseline during the first recovery period after isometric dorsi flexion, whereas unaffected subjects and patients with other causes of exercise-induced lower extremity pain reached their peak values during exercise (P<0.001). CONCLUSION: We have developed the first in-scanner MRI exercise protocol for the assessment of patients with suspected CECS. The technique shows high accuracy, sensitivity and specificity for diagnosis in this small cohort of patients with CECS. Further study may allow this non-invasive test to be used as a triage tool for invasive intracompartmental pressure measurements in patients with suspected CECS.