Dario Turco

Reliability of Total Renal Volume Computation in Polycystic Kidney Disease From Magnetic Resonance Imaging

RATIONALE AND OBJECTIVES Total renal volume (TRV) is an important quantitative indicator of the progression of autosomal dominant polycystic kidney disease (ADPKD). The Consortium for Radiologic Imaging Studies of Polycystic Kidney Disease proposes a method for TRV computation based on manual tracing and geometric modeling. Alternative approaches for TRV computation are represented by the application of advanced image processing techniques. In this study, we aimed to compare TRV estimates derived from these two different approaches. MATERIALS AND METHODS The nearly automated technique for the analysis of magnetic resonance (MR) images was tested on 30 ADPKD patients. TRV was computed from both axial (KVax) and coronal (KVcor) acquisitions and compared to measurements based on geometric modeling (KVap) by linear regression and Bland-Altman analysis. In addition, to assess reproducibility, intraobserver and interobserver variabilities were computed. RESULTS Linear regression analysis between KVax and KVcor resulted in an excellent correlation (KVax = 1KVcor - 0.78; r(2) = 0.997). Bland-Altman analysis showed a negligible bias and narrow limits of agreement (bias: -11.7 mL; SD: 54.3 mL). Similar results were obtained by comparison of volumes obtained applying the nearly automated method and the one based on geometric modeling (y = 0.98x + 75.9; r(2) = 0.99; bias: -53.7 mL; SD: 108.1 mL). Importantly, geometric modeling does not provide reliable TRV estimates in huge kidney affected by regional deformation. Intraobserver and interobserver variability resulted in very small percentage error <2%. CONCLUSIONS The results of this study provide the feasibility of using a nearly automated approach for accurate and fast evaluation of TRV also in markedly enlarged ADPKD kidneys including exophytic cysts.

Comparison of Total Kidney Volume Quantification Methods in Autosomal Dominant Polycystic Disease for a Comprehensive Disease Assessment

Background: In recent times, the scientific community has been showing increasing interest in the treatments aimed at slowing the progression of the autosomal dominant polycystic kidney disease (ADPKD). Therefore, in this paper, we test and evaluate the performance of several available methods for total kidney volume (TKV) computation in ADPKD patients - from echography to MRI - in order to optimize patient classification. Methods: Two methods based on geometric assumptions (mid-slice [MS], ellipsoid [EL]) and a third one on true contour detection were tested on 40 ADPKD patients at different disease stage using MRI. The EL method was also tested using ultrasound images in a subset of 14 patients. Their performance was compared against TKVs derived from reference manual segmentation of MR images. Patient clinical classification was also performed based on computed volumes. Results: Kidney volumes derived from echography significantly underestimated reference volumes. Geometric-based methods applied to MR images had similar acceptable results. The highly automated method showed better performance. Volume assessment was accurate and reproducible. Importantly, classification resulted in 79, 13, 10, and 2.5% of misclassification using kidney volumes obtained from echo and MRI applying the EL, the MS and the highly automated method respectively. Conclusion: Considering the fact that the image-based technique is the only approach providing a 3D patient-specific kidney model and allowing further analysis including cyst volume computation and monitoring disease progression, we suggest that geometric assumption (e.g., EL method) should be avoided. The contour-detection approach should be used for a reproducible and precise morphologic classification of the renal volume of ADPKD patients.

3D patient-specific models for left atrium characterization to support ablation in atrial fibrillation patients

BACKGROUND Radiofrequency ablation (RFA) is an important and promising therapy for atrial fibrillation (AF) patients. Optimization of patient selection and the availability of an accurate anatomical guide could improve RFA success rate. In this study we propose a unified, fully automated approach to build a 3D patient-specific left atrium (LA) model including pulmonary veins (PVs) in order to provide an accurate anatomical guide during RFA and without PVs in order to characterize LA volumetry and support patient selection for AF ablation. METHODS Magnetic resonance data from twenty-six patients referred for AF RFA were processed applying an edge-based level set approach guided by a phase-based edge detector to obtain the 3D LA model with PVs. An automated technique based on the shape diameter function was designed and applied to remove PVs and compute LA volume. 3D LA models were qualitatively compared with 3D LA surfaces acquired during the ablation procedure. An expert radiologist manually traced the LA on MR images twice. LA surfaces from the automatic approach and manual tracing were compared by mean surface-to-surface distance. In addition, LA volumes were compared with volumes from manual segmentation by linear and Bland-Altman analyses. RESULTS Qualitative comparison of 3D LA models showed several inaccuracies, in particular PVs reconstruction was not accurate and left atrial appendage was missing in the model obtained during RFA procedure. LA surfaces were very similar (mean surface-to-surface distance: 2.3±0.7mm). LA volumes were in excellent agreement (y=1.03x-1.4, r=0.99, bias=-1.37ml (-1.43%) SD=2.16ml (2.3%), mean percentage difference=1.3%±2.1%). CONCLUSIONS Results showed the proposed 3D patient-specific LA model with PVs is able to better describe LA anatomy compared to models derived from the navigation system, thus potentially improving electrograms and voltage information location and reducing fluoroscopic time during RFA. Quantitative assessment of LA volume derived from our 3D LA model without PVs is also accurate and may provide important information for patient selection for RFA.

Geometry-independent assessment of renal volume in polycystic kidney disease from magnetic resonance imaging

Total renal volume (TRV) is an important quantitative indicator of the progression of autosomal dominant polycystic kidney disease (ADPKD). The Consortium for Radiologic Imaging Studies of Polycystic Kidney Disease proposes a method for TRV computation based on manual tracing and geometric modeling. We developed a fast and nearly-automated technique for kidney segmentation and automatically compute TRV. In this study we aim to compare TRV estimates derived from these two different approaches. The highly-automated technique for the analysis of MR images was tested on 30 ADPKD patients. TRV was computed from both axial and coronal acquisitions, and compared to measurements based on geometric modeling by linear regression and Bland Altman analysis. In addition, to assess reproducibility, intra-observer and inter-observer variabilities were computed. The results of this study provide the feasibility of using a nearly-automated approach for accurate and fast evaluation of TRV also in markedly enlarged ADPKD kidneys.

Comment on the paper: Novel approach to estimate kidney and cyst volumes using mid-slice magnetic resonance images in polycystic kidney disease

idea the authors had that a simplification of this computation to speed up the measuring process could have an impact on the clinical scenario. Their solution is based on the analysis of a limited number of MR images to approximate kidney and cyst volumes. To this regard, we strongly believe that using image processing techniques to drastically reduce processing time and fully exploit the information contained in the acquired dataset without applying any geometrical approximation and provide reliable volume estimates could be helpful for the computation of the parameters CRISP proved to be clinically important. The availability of good quality MR images, refined image processing algorithms and computational power makes this option completely feasible. Indeed, some works proposing semiautomated or highly automated methods to quantify total kidney volumes in autosomal dominant polycystic kidney disease patients without using geometrical modeling and approximations have been published [2–4] . In particular, in a work that some of us coauthored [3] , an extremely strong correlation (r = 0.99) was found between an automated measurement (requiring about We recently read and appreciated the paper ‘Novel approach to estimate kidney and cyst volumes using mid-slice magnetic resonance images in polycystic kidney disease’ by Bae et al. [1] . The paper is very interesting and we recognize the intensive and very useful work the Consortium for Radiologic Imaging Studies of Polycystic Kidney Disease (CRISP) has conducted in the last years on this topic to develop methods to define a surrogate marker of the disease. Indeed, one of the most relevant results of CRISP was the evidence that magnetic resonance (MR) imaging is the best procedure for estimating changes in renal and cystic volume in autosomal dominant polycystic kidney disease over short followup periods. As stated in the study [1] , assessment of total renal and cystic volume is usually computed by manually tracing renal or cyst contours in a set of contiguous MR images and by summing the products of the area measurements and the slice thickness. We completely agree with the authors that measuring the area of every slice in such amounts of images is laborious, time-consuming and not applicable in routine clinical practice. For this reason we share the Published online: February 11, 2014

Fully Automated Segmentation of Polycystic Kidneys From Noncontrast Computed Tomography: A Feasibility Study and Preliminary Results

RATIONALE AND OBJECTIVES Total kidney volume is an important biomarker for the evaluation of autosomal dominant polycystic kidney disease progression. In this study, we present a novel approach for automated segmentation of polycystic kidneys from non-contrast-enhanced computed tomography (CT) images. MATERIALS AND METHODS Non-contrast-enhanced CT images were acquired from 21 patients with a diagnosis of autosomal dominant polycystic kidney disease. Kidney volumes obtained from the fully automated method were compared to volumes obtained by manual segmentation and evaluated using linear regression and Bland-Altman analyses. Dice coefficient was used for performance evaluation. RESULTS Kidney volumes from the automated method well correlated with the ones obtained by manual segmentation. Bland-Altman analysis showed a low percentage bias (-0.3%) and narrow limits of agreements (11.0%). The overlap between the three-dimensional kidney surfaces obtained with our approach and by manual tracing, expressed in terms of Dice coefficient, showed good agreement (0.91 ± 0.02). CONCLUSIONS This preliminary study showed the proposed fully automated method for renal volume assessment is feasible, exhibiting how a correct use of biomedical image processing may allow polycystic kidney segmentation also in non-contrast-enhanced CT. Further investigation on a larger dataset is needed to confirm the robustness of the presented approach.

Assessment of kidney volumes in polycystic kidney disease from coronal and axial MR images

Total renal volume (TRV) and individual renal cysts are important quantitative indicators of the progression of autosomal dominant polycystic kidney disease (ADPKD). TRV has been assessed by manually tracing renal contours from CT or MR scans, often employing contrast media (CM). We developed a fast and nearly automated technique based on the analysis of MR axial images acquired without CM injection for TRV quantification. Twenty ADPKD patients underwent MRI. Automatic extraction of kidney contours was performed on each acquired slice; the segmentation procedure was based on region growing and on the application of morphological operators and curvature-based motion. The area inside each contour was calculated and TRV was derived. Volume measurements were compared between axial and coronal acquisitions. TRV estimated in patients from axial acquisition was 1931±1117ml (range: 428÷4622ml) and from coronal acquisition volumes resulted in 1943±1135ml (range: 393÷4604ml) (p>0.05). These automatic measurements were in excellent correlation (r=0.99, y=1.01×-13.9) with a small bias and narrow limits of agreement in both absolute (-12±93ml) and percentage (-0.2±4.6%) terms. This preliminary study showed TRV from axial and coronal MRI acquisitions can be assessed automatically and with comparable and accurate results without requiring the use of potentially nephrotoxic contrast medium.

An exploratory study on coronary sinus lead tip three-dimensional trajectory changes in cardiac resynchronization therapy

BACKGROUND Prediction of response to cardiac resynchronization therapy (CRT) is still an unsolved major issue. The interface between left ventricular mechanics, coronary sinus (CS) lead, and pacing delivery has been little investigated. OBJECTIVE To investigate CS lead tip movements at baseline and during biventricular pacing (BiV) in the hypothesis that they could provide some insights into left ventricular mechanical behavior in CRT. METHODS Three-dimensional reconstruction of CS lead tip trajectory throughout the cardiac cycle using a novel fluoroscopy-based method was performed in 22 patients with chronic heart failure (19 men; mean age 70 ± 10 years). Three trajectories were computed: before (T-1) and immediately after (T0) BiV start-up and after 6 months (T1). CRT response was the echocardiographic end-systolic volume reduction ≥15% at T1. Metrics describing trajectory at T0, T-1, and T1 were compared between 9 responders (R) and 13 nonresponders (NR). RESULTS At T-1 trajectories demonstrated heterogeneous shapes and metrics, but at T0 the variations in the ratio between the two main axes (S1/S2) and in the eccentricity were statistically different between R and NR, pointing out a trajectorys change toward a significantly more circular shape at BiV start-up in R. Remarkably, R and NR could be completely separated by means of the percent variation in S1/S2 from T-1 to T0 (R: 47.5% [31.5% to 54.1%] vs. NR: -25.6% [-67% to -6.5%]). This single marker computed at T0 would have predicted CRT response at T1. CONCLUSIONS Preliminary data showed that CS lead tip trajectory changes induced by BiV were related to mechanical resynchronization.

Supervised learning modelization and segmentation of cardiac scar in delayed enhanced MRI

Delayed Enhancement Magnetic Resonance Imaging can be used to non-invasively differentiate viable from non-viable myocardium within the Left Ventricle in patients suffering from myocardial diseases. Automated segmentation of scarified tissue can be used to accurately quantify the percentage of myocardium affected. This paper presents a method for cardiac scar detection and segmentation based on supervised learning and level set segmentation. First, a model of the appearance of scar tissue is trained using a Support Vector Machines classifier on image-derived descriptors. Based on the areas detected by the classifier, an accurate segmentation is performed using a segmentation method based on level sets.