Imaging Markers Derived From MRI-Based Automated Kidney Segmentation—an Analysis of Data From the German National Cohort (NAKO Gesundheitsstudie)

BACKGROUND

Population-wide research on potential new imaging biomarkers of the kidney depends on accurate automated segmentation of the kidney and its compartments (cortex, medulla, and sinus).

METHODS

We developed a robust deep-learning framework for kidney (sub-)segmentation based on a hierarchical, three-dimensional convolutional neural network (CNN) that was optimized for multi-scale problems of combined localization and segmentation. We applied the CNN to abdominal magnetic resonance images from the population-based German National Cohort (NAKO) study.

RESULTS

There was good to excellent agreement between the model predictions and manual segmentations. The median values for the body-surface normalized total kidney, cortex, medulla, and sinus volumes of 9934 persons were 158, 115, 43, and 24 mL/m2. Distributions of these markers are provided both for the overall study population and for a subgroup of persons without kidney disease or any associated conditions. Multivariable adjusted regression analyses revealed that diabetes, male sex, and a higher estimated glomerular filtration rate (eGFR) are important predictors of higher total and cortical volumes. Each increase of eGFR by one unit (i.e., 1 mL/min per 1.73 m2 body surface area) was associated with a 0.98 mL/m2 increase in total kidney volume, and this association was significant. Volumes were lower in persons with eGFR-defined chronic kidney disease.

CONCLUSION

The extraction of image-based biomarkers through CNN-based renal sub-segmentation using data from a population-based study yields reliable results, forming a solid foundation for future investigations.

Longitudinal cardiac dimensions in patients undergoing LVAD implantation

BACKGROUND

In conventional left ventricular assist devices (LVAD), a separate outflow graft is sutured to the ascending aorta. Novel device designs may include a transventricular outflow cannula crossing the aortic valve (AV). While transversal ventricular dimensions are well investigated in patients with severe heart failure, little is known about the longitudinal dimensions. These dimensions are, however, particularly critical for the design and development of mechanical circulatory support (MCS) devices with transaortic outflow cannula.

METHODS

In an explorative retrospective cohort study at the University Medical Center Freiburg, Germany, the longitudinal cardiac dimensions of patients undergoing computed tomography angiography (CTA) before and, if available, after LVAD implantation were analyzed. Among others, the following dimensions were assessed: (a) apex to AV, (b) apex to mitral valve, (c) AV to sinotubular junction (STJ), (d) apex to STJ, (e) apex to brachiocephalic artery (BCA), and (f) AV to BCA.

RESULTS

In total, 44 LVAD patients (36 male, age 55.8 years, height 1.75 m) were included. The longitudinal cardiac dimensions were (a) 114.5 ± 12.1 mm, (b) 108.0 ± 12.4 mm, (c) 20.9 ± 2.9, (d) 135.4 ± 13.4 mm, (e) 206.0 ± 18.3, and (f) 91.5 ± 9.8 mm. Postoperatively, (a) and (b) decreased by 31.5% and 39.5%, respectively (N = 14).

CONCLUSIONS

Longitudinal cardiac dimensions may be reduced by up to 40% after LVAD implantation. A better knowledge of these dimensions and their postoperative alterations in LVAD patients may improve surgical planning and help to design MCS devices with transventricular outflow cannula.

Multiclass datasets expand neural network utility: an example on ankle radiographs

PURPOSE

Artificial intelligence in computer vision has been increasingly adapted in clinical application since the implementation of neural networks, potentially providing incremental information beyond the mere detection of pathology. As its algorithmic approach propagates input variation, neural networks could be used to identify and evaluate relevant image features. In this study, we introduce a basic dataset structure and demonstrate a pertaining use case.

METHODS

A multidimensional classification of ankle x-rays (n = 1493) rating a variety of features including fracture certainty was used to confirm its usability for separating input variations. We trained a customized neural network on the task of fracture detection using a state-of-the-art preprocessing and training protocol. By grouping the radiographs into subsets according to their image features, the influence of selected features on model performance was evaluated via selective training.

RESULTS

The models trained on our dataset outperformed most comparable models of current literature with an ROC AUC of 0.943. Excluding ankle x-rays with signs of surgery improved fracture classification performance (AUC 0.955), while limiting the training set to only healthy ankles with and without fracture had no consistent effect.

CONCLUSION

Using multiclass datasets and comparing model performance, we were able to demonstrate signs of surgery as a confounding factor, which, following elimination, improved our model. Also eliminating pathologies other than fracture in contrast had no effect on model performance, suggesting a beneficial influence of feature variability for robust model training. Thus, multiclass datasets allow for evaluation of distinct image features, deepening our understanding of pathology imaging.

MO061: Mri-Based Segmentation of Kidneys in Participants of the German National Cohort (NAKO/GNC) Study

BACKGROUND AND AIMS

Chronic kidney disease (CKD) affects around 10% of adults worldwide and an estimated 13–17% in Germany1. Imaging is a novel, promising approach to identify additional markers of kidney function and CKD2. Within a large, population-based cohort study, the German National Cohort study (NAKO/GNC), 30 000 participants underwent a whole-body MRI protocol3. The goal of our project was to develop an automated kidney segmentation workflow and to examine distributions of the segmented kidneys and kidney sub-compartments.

METHOD

Data from the first 11 207 participants were used to develop a robust image processing pipeline for kidney segmentation and apply it to participants’ abdominal MRI images. After importing 3D gradient echo and 2D haste images into the imaging platform NORA (www.nora-imaging.org), an in-house ‘patchwork’-framework (https://bitbucket.org/reisert/patchwork) based on deep-learning convolutional neural networks was trained on data from 300 persons to automatically segment different kidney compartments (cortex, hilus, medulla and cysts). After an initial training round, the model was improved over four iterations by a loop of prediction, manual correction and retraining. The final model was then applied to the full dataset of 11 207 abdominal MR images, followed by manual quality control prior to statistical analysis. Volumetric parameters for total kidney volume [TKV, defined as cortex + medulla] and compartments were calculated from the segmentations. Values were calculated in absolute units of mL and normalized to body-surface-area (BSA) defined as sqrt(weight in kg*height in cm)/3600.

RESULTS

TKV and the kidney compartments cortex, medulla and hilus could be segmented robustly with the trained network (Fig. 1A). After exclusion of approximately 10% of images because of insufficient segmentation quality due to initial imaging artifacts or poor image quality, the mean (SD) TKV was 364 (±60) mL for men and 290 (±51) mL for women. This difference was markedly attenuated after normalisation to BSA (Fig. 1B). The right kidney was systematically smaller than the left kidney by approximately 5% (Fig. 1C). There was a strong association between participants’ BSA with TKV (Fig. 1D). The normalized kidney sub-compartment volumes showed different patterns across age, with medullary volume decreasing and hilus increasing (Fig. 2).

CONCLUSION

The developed framework enables robust segmentation of kidneys in abdominal MRI data from a nonspecific clinical routine protocol of a large cohort study. Basic parameters such as TKV and sub-compartment volumes of the kidney show correlations with participants’ height, weight, sex and age that are consistent with prior knowledge and may enable an estimation of ‘kidney age’. This is an optimal starting point to identify more advanced imaging biomarkers of kidney function and CKD.

Determination of local flow ratios and velocities in a femoral venous cannula with computational fluid dynamics and 4D flow-sensitive magnetic resonance imaging: A method validation

Cannulas with multi-staged side holes are the method of choice for femoral cannulation in extracorporeal therapies today. A variety of differently designed products is available on the market. While the preferred tool for the performance assessment of such cannulas are pressure-flow curves, little is known about the flow and velocity distribution. Within this work flow and velocity patterns of a femoral venous cannula with multi-staged side holes were investigated. A mock circulation loop for cannula performance evaluation was built and reproduced using a computer-aided design system. With computational fluid dynamics, volume flows and fluid velocities were determined quantitatively and visually with hole-based precision. In order to ensure the correctness of the flow simulation, the results were subsequently validated by determining the same parameters with four-dimensional flow-sensitive magnetic resonance imaging. Measurement data and numerical solution differed 7% on average throughout the data set for the examined parameters. The highest inflow and velocity were detected at the most proximal holes, where half of the total volume flow enters the cannula. At every hole stage a Y-shaped inflow profile was detected, forming a centered stream in the middle of the cannula. Simultaneously, flow separation creates zones with significant lower flow velocities. Numerical simulation, validated with four-dimensional flow-sensitive magnetic resonance imaging, is a valuable tool to examine flow and velocity distributions of femoral venous cannulas with hole-based accuracy. Flow and velocity distribution in such cannulas are not ideal. Based on this work future cannulas can be effectively optimized.

Preclinical 4D-flow magnetic resonance phase contrast imaging of the murine aortic arch

PURPOSE

Cardiovascular diseases remain the number one death cause worldwide. Preclinical 4D flow phase contrast magnetic resonance imaging can provide substantial insights in the analysis of aortic pathophysiologies in various animal models. These insights may allow a better understanding of pathophysiologies, therapy monitoring, and can possibly be translated to humans. This study provides a framework to acquire the velocity field within the aortic arch. It analyses important flow values at different locations within the aortic arch. Imaging parameters with high temporal and spatial resolution are provided, that still allow combining this time-consuming method with other necessary imaging-protocols.

METHODS

A new setup was established where a prospectively gated 4D phase contrast sequence is combined with a highly sensitive cryogenic coil on a preclinical magnetic resonance scanner. The sequence was redesigned to maintain a close to steady state condition of the longitudinal magnetization and hence to overcome steady state artifacts. Imaging parameters were optimized to provide high spatial and temporal resolution. Pathline visualizations were generated from the acquired velocity data in order to display complex flow patterns.

RESULTS

Our setup allows data acquisition with at least two times the rate than that of previous publications based on Cartesian encoding, at an improved image quality. The "steady state" sequence reduces observed artifacts and provides uniform image intensity over the heart cycle. This made possible quantification of blood speed and wall shear stress (WSS) within the aorta and its branches. The highest velocities were observed in the ascending aorta with 137.5 ± 8 cm/s. Peak velocity values in the Brachiocephalic trunk were 57 ± 12 cm/s. Quantification showed that the peak flow occurs around 20 ms post R-wave in the ascending aorta. The highest mean axial wall shear stress was observed in the analysis plane between the left common carotid artery (LCCA) and the left subclavian artery. A stable image quality allows visualizing complex flow patterns by means of streamlines and for the first time, to the best of our knowledge, pathline visualizations from 4D flow MRI in mice.

CONCLUSION

The described setup allows analyzing pathophysiologies in mouse models of cardiovascular diseases in the aorta and its branches with better image quality and higher spatial and temporal resolution than previous Cartesian publications. Pathlines provide an advanced analysis of complex flow patterns in the murine aorta. An imaging protocol is provided that offers the possibility to acquire the aortic arch at sufficiently high resolution in less than one hour. This allows the combination of the flow assessment with other multifunctional imaging protocols.

Fenestrated and Branched Aortic Grafts

BACKGROUND

Abdominal and thoracic aortic aneurysms are diagnosed in 40 and 10 to 15 out of 100 000 persons per year, respectively. Fenestrated (fEVAR) and branched (bEVAR) stent grafts have been developed for abdominal juxtarenal and thoracoabdominal aneurysms. We discuss the patency and complication rates of fEVAR and bEVAR procedures and compare them with the outcome of open surgery.

METHODS

This review is based on pertinent publications from 2011 to 2014 that were retrieved by a selective literature search. The clinical outcomes of case series involving a total of more than 1500 patients are presented. The discussion takes account of recommendations contained in the literature and the authors' own experience.

RESULTS

Open surgery and aortic stent grafting have not been compared in any randomized trial to date. We identified 7 clinical series that included a total of 1270 fEVAR patients and 5 with a total of 408 bEVAR patients. The perioperative mortality after fEVAR procedures was 0-4%. Spinal cord ischemia arose in 1% of cases. The stent patency rate in visceral vessels ranged from 93 to 98%. bEVAR procedures were associated with both higher mortality (4-7%) and more common spinal cord ischemia (4-13%). 5-8% of all patients needed dialysis perioperatively, and the stent patency rate in visceral vessels was 94-97%. Preoperative renal insufficiency was a risk factor for peri-interventional death. Impaired renal function after fEVAR/bEVAR procedures was mainly associated with intermittent lower limb ischemia.

CONCLUSION

The results of fEVAR/bEVAR procedures in the last 5 years are similar to those of open surgery. The high postoperative rate of spinal cord ischemia remains a serious problem in the endovascular treatment of thoracoabdominal aortic aneurysms. The decision to implant a stent graft by an endovascular approach or to treat surgically should be made on a case-to-case basis in an interdisciplinary vascular conference.

K-t GRAPPA accelerated phase contrast MRI: Improved assessment of blood flow and 3-directional myocardial motion during breath-hold

PURPOSE

To evaluate spatiotemporal parallel imaging with R = 5 in comparison to conventional parallel imaging with R = 2 applied to phase contrast (PC) magnetic resonance imaging (MRI). This was motivated by the fact that scan times for PC imaging often exceed breath-hold capabilities of patients even with standard parallel imaging using typical reduction factors of R = 2.

MATERIALS AND METHODS

K-t generalized autocalibrating partially parallel acquisition (GRAPPA) acquisition was validated in phantom measurements and then applied in 10 volunteer and three patient examinations. Due to the higher reduction factor compared to conventional GRAPPA, k-t GRAPPA measurements could be performed during breath-hold with high spatial and temporal resolution. K-t GRAPPA scans were compared to GRAPPA acquired during free-breathing with navigator respiration control. In addition, spatiotemporally accelerated PC imaging was acquired during free-breathing for comparison of k-t-accelerated breath-held scans.

RESULTS

Substantial improvements in image quality for the breath-hold measurements were observed. Significantly reduced peak velocities were found for the GRAPPA protocol compared to the k-t-accelerated breath-hold scans for both flow (8%) and myocardial motion (up to 30%) measurements.

CONCLUSION

Spatiotemporal acceleration allows the performance of high temporal or spatial resolution PC imaging during breath-hold while providing high image quality and robust acquisition of functional information that cannot be achieved during breath-hold with standard techniques.

Evaluation of 3D blood flow patterns and wall shear stress in the normal and dilated thoracic aorta using flow-sensitive 4D CMR

BACKGROUND

The purpose of this study was to investigate 3D flow patterns and vessel wall parameters in patients with dilated ascending aorta, age-matched subjects, and healthy volunteers.

METHODS

Thoracic time-resolved 3D phase contrast CMR with 3-directional velocity encoding was applied to 33 patients with dilated ascending aorta (diameter≥40 mm, age=60±16 years), 15 age-matched normal controls (diameter≤37 mm, age=68±7.5 years) and 15 young healthy volunteers (diameter≤30 mm, age=23±2 years). 3D blood flow was visualized and flow patterns were graded regarding presence of supra-physiologic-helix and vortex flow using a semi-quantitative 3-point grading scale. Blood flow velocities, regional wall shear stress (WSS), and oscillatory shear index (OSI) were quantified.

RESULTS

Incidence and strength of supra-physiologic-helix and vortex flow in the ascending aorta (AAo) was significantly higher in patients with dilated AAo (16/33 and 31/33, grade 0.9±1.0 and 1.5±0.6) than in controls (2/15 and 7/15, grade 0.2±0.6 and 0.6±0.7, P<.05) or healthy volunteers (1/15 and 0/15, grade 0.1±0.3 P<.05). Greater strength of the ascending aortic helix and vortex flow were associated with significant differences in AAo diameters (P<.05). Peak systolic WSS in the ascending aorta and aortic arch was significantly lower in patients with dilated AAo (P<.0157-.0488). AAo diameter positively correlated to time to peak systolic velocities (r=0.30-0.53, P<.04), OSI (r=0.33-0.49, P<0.02) and inversely correlated to peak systolic WSS (r=0.32-0.40, P<.03). Peak systolic WSS was significantly lower in AAo aneurysms at the right and outer curvature within the AAo and proximal arch (P<.01-.05).

CONCLUSIONS

Increase in AAo diameter is significantly correlated with the presence and strength of supra-physiologic-helix and vortex formation in the AAo, as well with decrease in systolic WSS and increase in OSI.

MR-based visualization and quantification of three-dimensional flow characteristics in the portal venous system

PURPOSE

To evaluate the feasibility of time-resolved flow-sensitive MRI for the three-dimensional (3D) visualization and quantification of normal and pathological portal venous (PV) hemodynamics.

MATERIALS AND METHODS

Portal venous hemodynamics were evaluated in 18 healthy volunteers and 5 patients with liver cirrhosis. ECG- and adaptive respiratory navigator gated flow-sensitive 4D MRI (time-resolved 3D MRI with three-directional velocity encoding) was performed on a 3 Tesla MR system (TRIO, Siemens, Germany). Qualitative flow analysis was achieved using 3D streamlines and time-resolved particle traces originating from seven emitter planes precisely placed at anatomical landmarks in the PV system. Quantitative analysis included retrospective extraction of regional peak and mean velocities and vessel area. Results were compared with standard 2D flow-sensitive MRI and to the reference standard Doppler ultrasound.

RESULTS

Qualitative flow analysis was successfully used in the entire PV system. Venous hemodynamics in all major branches in 17 of 18 volunteers and 3 of 5 patients were reliably depicted with good interobserver agreement (kappa = 0.62). Quantitative analysis revealed no significant differences and moderate agreement for peak velocities between 3D MR and 2D MRI (r = 0.46) and Doppler ultrasound (US) (r = 0.35) and for mean velocities between 3D and 2D MRI (r = 0.41). The PV area was significantly (P < 0.01) higher in 3D and 2D MRI compared with US.

CONCLUSION

We successfully applied 3D MR velocity mapping in the PV system, providing a detailed qualitative and quantitative analysis of normal and pathological hemodynamics.