Reconstruction of 3D knee MRI using deep learning and compressed sensing: a validation study on healthy volunteers.

BACKGROUND: To investigate the potential of combining compressed sensing (CS) and artificial intelligence (AI), in particular deep learning (DL), for accelerating three-dimensional (3D) magnetic resonance imaging (MRI) sequences of the knee. METHODS: Twenty healthy volunteers were examined using a 3-T scanner with a fat-saturated 3D proton density sequence with four different acceleration levels (10, 13, 15, and 17). All sequences were accelerated with CS and reconstructed using the conventional and a new DL-based algorithm (CS-AI). Subjective image quality was evaluated by two blinded readers using seven criteria on a 5-point-Likert-scale (overall impression, artifacts, delineation of the anterior cruciate ligament, posterior cruciate ligament, menisci, cartilage, and bone). Using mixed models, all CS-AI sequences were compared to the clinical standard (sense sequence with an acceleration factor of 2) and CS sequences with the same acceleration factor. RESULTS: 3D sequences reconstructed with CS-AI achieved significantly better values for subjective image quality compared to sequences reconstructed with CS with the same acceleration factor (p ≤ 0.001). The images reconstructed with CS-AI showed that tenfold acceleration may be feasible without significant loss of quality when compared to the reference sequence (p ≥ 0.999). CONCLUSIONS: For 3-T 3D-MRI of the knee, a DL-based algorithm allowed for additional acceleration of acquisition times compared to the conventional approach. This study, however, is limited by its small sample size and inclusion of only healthy volunteers, indicating the need for further research with a more diverse and larger sample. TRIAL REGISTRATION: DRKS00024156. RELEVANCE STATEMENT: Using a DL-based algorithm, 54% faster image acquisition (178 s versus 384 s) for 3D-sequences may be possible for 3-T MRI of the knee. KEY POINTS: • Combination of compressed sensing and DL improved image quality and allows for significant acceleration of 3D knee MRI. • DL-based algorithm achieved better subjective image quality than conventional compressed sensing. • For 3D knee MRI at 3 T, 54% faster image acquisition may be possible.

Reconstruction of shoulder MRI using deep learning and compressed sensing: a validation study on healthy volunteers.

BACKGROUND: To investigate the potential of combining compressed sensing (CS) and deep learning (DL) for accelerated two-dimensional (2D) and three-dimensional (3D) magnetic resonance imaging (MRI) of the shoulder. METHODS: Twenty healthy volunteers were examined using at 3-T scanner with a fat-saturated, coronal, 2D proton density-weighted sequence with four acceleration levels (2.3, 4, 6, and 8) and a 3D sequence with three acceleration levels (8, 10, and 13), all accelerated with CS and reconstructed using the conventional algorithm and a new DL-based algorithm (CS-AI). Subjective image quality was evaluated by two blinded readers using 6 criteria on a 5-point Likert scale (overall impression, artifacts, and delineation of the subscapularis tendon, bone, acromioclavicular joint, and glenoid labrum). Objective image quality was measured by calculating signal-to-noise-ratio, contrast-to-noise-ratio, and a structural similarity index measure. All reconstructions were compared to the clinical standard (CS 2D acceleration factor 2.3; CS 3D acceleration factor 8). Additionally, subjective and objective image quality were compared between CS and CS-AI with the same acceleration levels. RESULTS: Both 2D and 3D sequences reconstructed with CS-AI achieved on average significantly better subjective and objective image quality compared to sequences reconstructed with CS with the same acceleration factor (p ≤ 0.011). Comparing CS-AI to the reference sequences showed that 4-fold acceleration for 2D sequences and 13-fold acceleration for 3D sequences without significant loss of quality (p ≥ 0.058). CONCLUSIONS: For MRI of the shoulder at 3 T, a DL-based algorithm allowed additional acceleration of acquisition times compared to the conventional approach. RELEVANCE STATEMENT: The combination of deep-learning and compressed sensing hold the potential for further scan time reduction in 2D and 3D imaging of the shoulder while providing overall better objective and subjective image quality compared to the conventional approach. TRIAL REGISTRATION: DRKS00024156. KEY POINTS: • Combination of compressed sensing and deep learning improved image quality and allows for significant acceleration of shoulder MRI. • Deep learning-based algorithm achieved better subjective and objective image quality than conventional compressed sensing. • For shoulder MRI at 3 T, 40% faster image acquisition for 2D sequences and 38% faster image acquisition for 3D sequences may be possible.

Speeding up the clinical routine: Compressed sensing for 2D imaging of lumbar spine disc herniation.

PURPOSE: Increasing economic pressure and patient demands for comfort require an ever-increasing acceleration of scan times without compromising diagnostic certainty. This study tested the new acceleration technique Compressed SENSE (CS-SENSE) as well as different reconstruction methods for the lumbar spine. METHODS: In this prospective study, 10 volunteers and 14 patients with lumbar disc herniation were scanned using a sagittal 2D T2 turbo spin echo (TSE) sequence applying different acceleration factors of SENSE and CS-SENSE. Gradient echo (GRE), autocalibration (CS-Auto) and TSE prescans were tested for reconstruction. Images were analysed by two readers regarding anatomical delineation, diagnostic certainty (for patients only) and image quality as well as objectively calculating the root mean square error (RMSE), structural similarity index (SSIM), SNR and CNR. The Friedman test and Chi-squared were used for ordinal, ANOVA for repeated measurements and Tukey Kramer test for continuous data. Cohen's kappawas calculated for interreader reliability. RESULTS: CS-SENSE outperformed SENSE and CS-Auto regarding RMSE (e.g. CS-SENSE 1.5: 43.03 ±â€¯11.64 versus SENSE 1.5: 80.41 ±â€¯17.66; p = 0.0038) and SSIM as well as in the subjective rating for CS-SENSE 3 TSE. In the patient setting image quality was unchanged in all subjective criteria up to CS-SENSE 3 TSE (all p > 0.05) compared to standard T2 with 43 % less scan time while the GRE prescan only allowed a reduction of 32 %. CONCLUSION: Combining a TSE prescan with CS-SENSE enables significant scan time reductions with unchanged ratings for lumbar spine disc herniation making this superior to the currently used SENSE acceleration or GRE reconstructions.

Lymph Node Assessment in Prostate Cancer: Evaluation of Iodine Quantification With Spectral Detector CT in Correlation to PSMA PET/CT.

PURPOSE: The aims of this study were to evaluate spectral detector CT (SDCT)-derived iodine concentration (IC) of lymph nodes diagnosed as metastatic and benign in prostate-specific membrane antigen (PSMA) PET/CT and to assess its potential use for lymph node assessment in prostate cancer. PATIENTS AND METHODS: Thirty-four prostate cancer patients were retrospectively included: 16 patients with and 18 without lymph node metastases as determined by PSMA PET/CT. Patients underwent PSMA PET/CT as well as portal venous phase abdominal SDCT for clinical cancer follow-up. Only scan pairs with a stable nodal status indicated by constant size as well as comparable prostate-specific antigen (PSA) levels were included. One hundred benign and 96 suspected metastatic lymph nodes were annotated and correlated between SDCT and PSMA PET/CT. Iodine concentration in SDCT-derived iodine maps and SUVmax in ultra-high definition reconstructions from PSMA PET/CT were acquired based on the region of interest. RESULTS: Metastatic lymph nodes as per PSMA PET/CT showed higher IC than nonmetastatic nodes (1.9 ± 0.6 mg/mL vs 1.5 ± 0.5 mg/mL, P < 0.05) resulting in an AUC of 0.72 and sensitivity/specificity of 81.3%/58.5%. The mean short axis diameter of metastatic lymph nodes was larger than that of nonmetastatic nodes (6.9 ± 3.6 mm vs 5.3 ± 1.3 mm; P < 0.05); a size threshold of 1 cm short axis diameter resulted in a sensitivity/specificity of 12.8%/99.0%. There was a significant yet weak positive correlation between SUVmax and IC (rs = 0.25; P < 0.001). CONCLUSIONS: Spectral detector CT-derived IC was increased in lymph nodes diagnosed as metastatic in PSMA PET/CT yet showed considerable data overlap. The correlation between IC and SUVmax was weak, highlighting the role of PSMA PET/CT as important reference imaging modality for detection of lymph node metastases in prostate cancer patients.

Virtual monoenergetic images preserve diagnostic assessability in contrast media reduced abdominal spectral detector CT.

OBJECTIVES: To investigate if low-keV virtual monoenergetic images (VMI40keV) from abdominal spectral detector CT (SDCT) with reduced intravenous contrast media application (RCM) provide abdominal assessment similar to conventional images with standard contrast media (SCM) dose. METHODS: 78 patients with abdominal SDCT were retrospectively included: 41 patients at risk for adverse reactions who received 44 RCM examinations with 50 ml and 37 patients who underwent 44 SCM examinations with 100 ml of contrast media (CM) and who were matched for effective body diameters. RCM, SCM images and RCM-VMI40keV were reconstructed. Attenuation and signal-to-noise ratio (SNR) of liver, pancreas, kidneys, lymph nodes, psoas muscle, aorta and portal vein were assessed ROIs-based. Contrast-to-noise ratios (CNR) of lymph nodes vs aorta/portal vein were calculated. Two readers evaluated organ/vessel contrast, lymph node delineation, image noise and overall assessability using 4-point Likert scales. RESULTS: RCM were inferior to SCM images in all quantitative/qualitative criteria. RCM-VMI40keV and SCM images showed similar lymph node and muscle attenuation (p = 0.83,0.17), while for all other ROIs, RCM-VMI40keV showed higher attenuation (p ≤ 0.05). SNR was comparable between RCM-VMI40keV and SCM images (p range: 0.23-0.99). CNR of lymph nodes was highest in RCM-VMI40keV (p ≤ 0.05). RCM-VMI40keV received equivalent or higher scores than SCM in all criteria except for organ contrast, overall assessability and image noise, where SCM were superior (p ≤ 0.05). However, RCM-VMI40keV received proper or excellent scores in 88.6/94.2/95.4% of the referring cases. CONCLUSIONS: VMI40keV counteract contrast deterioration in CM reduced abdominal SDCT, facilitating diagnostic assessment. ADVANCES IN KNOWLEDGE: SDCT-derived VMI40keV provide adequate depiction of vessels, organs and lymph nodes even at notable CM reduction.

MRI Follow-up of Astrocytoma: Automated Coregistration and Color-Coding of FLAIR Sequences Improves Diagnostic Accuracy With Comparable Reading Time.

BACKGROUND: MRI follow-up is widely used for longitudinal assessment of astrocytoma, yet reading can be tedious and error-prone, in particular when changes are subtle. PURPOSE/HYPOTHESIS: To determine the effect of automated, color-coded coregistration (AC) of fluid attenuated inversion recovery (FLAIR) sequences on diagnostic accuracy, certainty, and reading time compared to conventional follow-up MRI assessment of astrocytoma patients. STUDY TYPE: Retrospective. POPULATION: In all, 41 patients with neuropathologically confirmed astrocytoma. FIELD STRENGTH/SEQUENCE: 1.0-3.0T/FLAIR ASSESSMENT: The presence or absence of tumor progression was determined based on FLAIR sequences, contrast-enhanced T1 sequences, and clinical data. Three radiologists assessed 47 MRI study pairs in a conventional reading (CR) and in a second reading supported by AC after 6 weeks. Readers determined the presence/absence of tumor progression and indicated diagnostic certainty on a 5-point Likert scale. Reading time was recorded by an independent assessor. STATISTICAL TESTS: The Wilcoxon test was used to assess reading time and diagnostic certainty. Differences in diagnostic accuracy, sensitivity, and specificity were analyzed with the McNemar mid-p test. RESULTS: Readers attained significantly higher overall sensitivity (0.86 vs. 0.75; P < 0.05) and diagnostic accuracy (0.84 vs. 0.73; P < 0.05) for detection of progressive nonenhancing tumor burden when using AC compared to CR. There was a strong trend towards higher specificity within the AC-augmented reading, yet without statistical significance (0.83 vs. 0.71; P = 0.08). Sensitivity for unequivocal disease progression was similarly high in both approaches (AC: 0.94, CR: 0.92), while for marginal disease progressions, it was significantly higher in AC (AC: 0.78, CR: 0.58; P < 0.05). Reading time including application loading time was comparable (AC: 38.1 ± 16.8 sec, CR: 36.0 ± 18.9 s; P = 0.25). DATA CONCLUSION: Compared to CR, AC improves comparison of FLAIR signal hyperintensity at MRI follow-up of astrocytoma patients, allowing for a significantly higher diagnostic accuracy, particularly for subtle disease progression at a comparable reading time. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY STAGE: 6 J. Magn. Reson. Imaging 2020;52:1197-1206.

Vasoconstriction and Impairment of Neurovascular Coupling after Subarachnoid Hemorrhage: a Descriptive Analysis of Retinal Changes.

Impaired cerebral autoregulation and neurovascular coupling (NVC) contribute to delayed cerebral ischemia after subarachnoid hemorrhage (SAH). Retinal vessel analysis (RVA) allows non-invasive assessment of vessel dimension and NVC hereby demonstrating a predictive value in the context of various neurovascular diseases. Using RVA as a translational approach, we aimed to assess the retinal vessels in patients with SAH. RVA was performed prospectively in 24 patients with acute SAH (group A: day 5-14), in 11 patients 3 months after ictus (group B: day 90 ± 35), and in 35 age-matched healthy controls (group C). Data was acquired using a Retinal Vessel Analyzer (Imedos Systems UG, Jena) for examination of retinal vessel dimension and NVC using flicker-light excitation. Diameter of retinal vessels-central retinal arteriolar and venular equivalent-was significantly reduced in the acute phase (p < 0.001) with gradual improvement in group B (p < 0.05). Arterial NVC of group A was significantly impaired with diminished dilatation (p < 0.001) and reduced area under the curve (p < 0.01) when compared to group C. Group B showed persistent prolonged latency of arterial dilation (p < 0.05). Venous NVC was significantly delayed after SAH compared to group C (A p < 0.001; B p < 0.05). To our knowledge, this is the first clinical study to document retinal vasoconstriction and impairment of NVC in patients with SAH. Using non-invasive RVA as a translational approach, characteristic patterns of compromise were detected for the arterial and venous compartment of the neurovascular unit in a time-dependent fashion. Recruitment will continue to facilitate a correlation analysis with clinical course and outcome.