Generalizability of Deep Learning Segmentation Algorithms for Automated Assessment of Cartilage Morphology and MRI Relaxometry.

BACKGROUND: Deep learning (DL)-based automatic segmentation models can expedite manual segmentation yet require resource-intensive fine-tuning before deployment on new datasets. The generalizability of DL methods to new datasets without fine-tuning is not well characterized. PURPOSE: Evaluate the generalizability of DL-based models by deploying pretrained models on independent datasets varying by MR scanner, acquisition parameters, and subject population. STUDY TYPE: Retrospective based on prospectively acquired data. POPULATION: Overall test dataset: 59 subjects (26 females); Study 1: 5 healthy subjects (zero females), Study 2: 8 healthy subjects (eight females), Study 3: 10 subjects with osteoarthritis (eight females), Study 4: 36 subjects with various knee pathology (10 females). FIELD STRENGTH/SEQUENCE: A 3-T, quantitative double-echo steady state (qDESS). ASSESSMENT: Four annotators manually segmented knee cartilage. Each reader segmented one of four qDESS datasets in the test dataset. Two DL models, one trained on qDESS data and another on Osteoarthritis Initiative (OAI)-DESS data, were assessed. Manual and automatic segmentations were compared by quantifying variations in segmentation accuracy, volume, and T2 relaxation times for superficial and deep cartilage. STATISTICAL TESTS: Dice similarity coefficient (DSC) for segmentation accuracy. Lin's concordance correlation coefficient (CCC), Wilcoxon rank-sum tests, root-mean-squared error-coefficient-of-variation to quantify manual vs. automatic T2 and volume variations. Bland-Altman plots for manual vs. automatic T2 agreement. A P value < 0.05 was considered statistically significant. RESULTS: DSCs for the qDESS-trained model, 0.79-0.93, were higher than those for the OAI-DESS-trained model, 0.59-0.79. T2 and volume CCCs for the qDESS-trained model, 0.75-0.98 and 0.47-0.95, were higher than respective CCCs for the OAI-DESS-trained model, 0.35-0.90 and 0.13-0.84. Bland-Altman 95% limits of agreement for superficial and deep cartilage T2 were lower for the qDESS-trained model, ±2.4 msec and ±4.0 msec, than the OAI-DESS-trained model, ±4.4 msec and ±5.2 msec. DATA CONCLUSION: The qDESS-trained model may generalize well to independent qDESS datasets regardless of MR scanner, acquisition parameters, and subject population. EVIDENCE LEVEL: 1 TECHNICAL EFFICACY: Stage 1.

Free-breathing magnetic resonance imaging with radial k-space sampling for neonates and infants to reduce anesthesia.

BACKGROUND: Conventional chest and abdominal MRI require breath-holds to reduce motion artifacts. Neonates and infants require general anesthesia with intubation to enable breath-held acquisitions. OBJECTIVE: We aimed to validate a free-breathing approach to reduce general anesthesia using a motion-insensitive radial acquisition with respiratory gating. MATERIALS AND METHODS: We retrospectively enrolled children <3 years old who were referred for MRI of the chest or abdomen. They were divided into two groups according to MRI protocol: (1) breath-held scans under general anesthesia with T2-weighted single-shot fast spin-echo (SSFSE) and contrast-enhanced T1-weighted modified Dixon, and (2) free-breathing scans using radial sequences (T2-W MultiVane XD and contrast-enhanced T1-W three-dimensional [3-D] Vane XD). Two readers graded image quality and motion artifacts. RESULTS: We included 23 studies in the free-breathing cohort and 22 in the breath-hold cohort. The overall imaging scores for the free-breathing radial T2-W sequence were similar to the scores for the breath-held T2-W SSFSE sequence (chest, 3.6 vs. 3.2, P=0.07; abdomen, 3.9 vs. 3.7, P=0.66). The free-breathing 3-D radial T1-W sequence also had image quality scores that were similar to the breath-held T1-W sequence (chest, 4.0 vs. 3.0, P=0.06; abdomen, 3.7 vs. 3.9, P=0.15). Increased motion was seen in the abdomen on the radial T2-W sequence (P<0.001), but increased motion was not different in the chest (P=0.73) or in contrast-enhanced T1-W sequences (chest, P=0.39; abdomen, P=0.15). The mean total sequence time was longer in free-breathing compared to breath-held exams (P<0.01); however, this did not translate to longer overall exam times (P=0.94). CONCLUSION: Motion-insensitive radial sequences used for infants and neonates were of similar image quality to breath-held sequences and had decreased sedation and intubation.

Attention-guided deep learning for gestational age prediction using fetal brain MRI.

Magnetic resonance imaging offers unrivaled visualization of the fetal brain, forming the basis for establishing age-specific morphologic milestones. However, gauging age-appropriate neural development remains a difficult task due to the constantly changing appearance of the fetal brain, variable image quality, and frequent motion artifacts. Here we present an end-to-end, attention-guided deep learning model that predicts gestational age with R2 score of 0.945, mean absolute error of 6.7 days, and concordance correlation coefficient of 0.970. The convolutional neural network was trained on a heterogeneous dataset of 741 developmentally normal fetal brain images ranging from 19 to 39 weeks in gestational age. We also demonstrate model performance and generalizability using independent datasets from four academic institutions across the U.S. and Turkey with R2 scores of 0.81-0.90 after minimal fine-tuning. The proposed regression algorithm provides an automated machine-enabled tool with the potential to better characterize in utero neurodevelopment and guide real-time gestational age estimation after the first trimester.

Use of compressed sensing to reduce scan time and breath-holding for cardiac cine balanced steady-state free precession magnetic resonance imaging in children and young adults.

BACKGROUND: Conventional pediatric volumetric MRI acquisitions of a short-axis stack typically require multiple breath-holds under anesthesia. OBJECTIVE: Here, we aimed to validate a vendor-optimized compressed-sensing approach to reduce scan time during short-axis balanced steady-state free precession (bSSFP) cine imaging. MATERIALS AND METHODS: Imaging was performed in 28 patients (16±9 years) in this study on a commercial 3-tesla (T) scanner using retrospective electrocardiogram-gated cine bSSFP. Cine short-axis images covering both ventricles were acquired with conventional parallel imaging and a vendor-optimized parallel imaging/compressed-sensing approach. Qualitative Likert scoring for blood-myocardial contrast, edge definition, and presence of artifact was performed by two experienced radiologists. Quantitative comparisons were performed including biventricular size and function. A paired t-test was used to detect significant differences (P<0.05). RESULTS: Scan duration was 7±2 s/slice for conventional imaging (147±33 s total) vs. 4±2 s/slice for compressed sensing (83±28 s total). No significant differences were found with qualitative image scores for blood-myocardial contrast, edge definition, and presence of artifact. No significant differences were found in volumetric analysis between the two sequences. The number of breath-holds was 10±4 for conventional imaging and 5±3 for compressed sensing. CONCLUSION: Compressed sensing allowed for a 50% reduction in the number of breath-holds and a 43% reduction in the total scan time without differences in the qualitative or quantitative measurements as compared to the conventional technique.

Significado do ponto de inflexäo inferior da curva pressão-volume em pacientes com insuficiência respiratória aguda: avaliaçäo por tomografia computadorizada / Significance of lower inflection point detected on lung pressure-volume: an assessment by computed tomography

O objetivo deste estudo foi avaliar, através de tomografia computadorizada, a morfologia pulmonar em pacientes com lesão pulmonar aguda, de acordo com a presença ou ausência de ponto de inflexão inferior (Pinf) nas curvas pressão-volume e comparar os efeitos da pressão expiratória final positiva (PEEP)...

Avaliaçäo por tomografia computadorizada da hiperdistensão pulmonar induzida por PEEP em indivíduos normais e em pacientes com insuficiência respiratória aguda / Computed tomography assessment of positive end-expiratory pressure-induced lung overdistension

O objetivo do presente estudo foi determinar o limite tomográfico da hiperdistensão pulmonar em indivíduos normais, bem como avaliar o recrutamento e a hiperdistensão pulmonares induzidos pela pressão expiratória final positiva em pacientes com lesão pulmonar aguda...

Comparaçäo de três métodos para medida das curvas de complacência em pacientes com insuficiência respiratória aguda / Comparison of three methods for measuring pressure-volume curves in patients with acute respiratory failure

As medidas da complacência respiratória a partir das curvas pressão-volume são indicadas para avaliar a gravidade da insuficiência respiratória aguda. O objetivo do presente estudo foi comparar diferentes métodos de obtenção das curvas pressão-volume e avaliar sua reprodutibilidade e fidedignidade...