5th Dutch Bio-Medical Engineering Conference 2015
22-23 January 2015, Egmond aan Zee, The Netherlands
13:30   Imaging II
15 mins
Olena Filatova, Hanne Kause, Andrea Fuster, Henk Marquering, Luc Florack, Hans van Assen
Abstract: For patients who require an aortic valve replacement, but have a high operative risk due to their age or other factors, the Transcatheter Aortic Valve Implantation (TAVI) technique has recently become available. Pre- and post-procedure imaging is performed in order to assess the patient-specific situation and to verify the success of valve implantation, often including tagging magnetic resonance imaging (tMRI). However, easy to use and efficient tools for the analysis of tMRI data are lacking. The goal of this project is to develop a user-friendly software tool that incorporates new tMRI image analysis techniques being developed at TU/e, and suitable for a clinical research environment. Consequently, clinical partners will be able to participate in the evaluation and improvement of the algorithms. In order to specify the requirements for the final software tool, several interviews have been done with both scientists and clinicians. Based on the acquired information, we chose to develop the software tool as a plug-in to existing multi-platform medical image processing and visualization software, called medInria (cf. http://med.inria.fr/). MedInria is a platform under active development, but it already allows file import of relevant formats (e.g. DICOM) and database management. The plug-in user interface supports the drag-and-drop feature as an easy-to-learn technique for choosing an input image or sequence from a database. An optic flow method [1] for tMRI analysis was implemented in C++. Microsoft Visual Studio 2010 was used as an integrated development environment with CMake 2.8 employed to manage the build process of the whole project depending on multiple libraries such as ITK, dTK and Qt. A stand-alone tMRI analysis software tool will provide the possibility to perform more clinically oriented research regarding cardiac function in the context of therapy selection, planning, and evaluation. The next steps for this tool include strain calculations based on the obtained optic flow velocities [2] and visualization of the results in bull’s eye plots. REFERENCES [1] H.C. van Assen et al., “Cardiac Strain and Rotation Analysis using Multi-scale Optical Flow," Computational Biomechanics for Medicine: Soft Tissues and the Muskulosceletal System, pp.91-102, 2011. [2] L. Florack and H. van Assen, “A new methodology for multiscale myocardial deformation and strain analysis based on tagging MRI,” International Journal of Biomedical Imaging, 2010.
15 mins
Sabine Krueger-Ziolek, Benjamin Schullcke, Knut Moeller
Abstract: Electrical impedance tomography (EIT) is a radiation-free imaging method which is applied to visualize regional lung dynamics. Regional changes of electrical impedance of lung tissue conditioned by varying amounts of gas volume during respiration can be measured and recorded. Respiration can be classified in two types: thoracic and abdominal breathing. Normal ventilation, which employs all lung regions, is a combination of both breathing types. In the following experiment, we investigated the impact of thoracic and abdominal breathing on EIT imaging in various thorax sections. EIT measurements were performed on two volunteers in three different chest planes (3rd, 5th and 7th intercostal space) during thoracic and abdominal tidal breathing. An EIT system with an electrode belt with 16 electrodes (PulmoVista 500, Dräger Medical, Germany) was applied. Functional residual capacity (FRC) was ascertained simultaneously using a special designed body plethysmograph. Variation of global impedance at end-expiration during tidal breathing for thoracic and abdominal breathing was determined in each thorax plane. FRC varied less than 2 % within all measurements for each volunteer, pointing to comparable end-expiratory levels. However, both volunteers showed a decrease by approx. 18 % in global impedance at end-expiration (ΔIFRC) between thoracic and abdominal breathing in the upper thorax section. In the middle thoracic plane differences in ΔIFRC between both respiration types were less than 5 %. A decline by 20 to 30 % in ΔIFRC from thoracic to abdominal breathing was visible in the lower chest plane. Results of our experiment indicated that global impedance measured at end-expiration during tidal breathing varied by type of respiration within electrode planes, although global FRC values differed rarely. We suspected that ΔIFRC variations in the upper thorax plane were caused by lung tissue distensions. We assumed that respiratory muscles might not be totally relaxed at point of end-expiration. Varying ΔIFRC values in the lower chest plane were probably caused by different positions of the diaphragm at end-expiration, which in turn led to the gathering of various amounts of lung tissue. In conclusion, changes in breathing type during EIT data collection probably leads to misinterpretation because of varying amounts of lung tissue, potentially differently ventilated, captured within electrode planes. Data measured in the upper and lower thorax plane appeared to be more error-prone than data collected in the middle chest plane. Respiration types seem to have different impact on EIT data, and thus have to be considered in data interpretation.
15 mins
Salvatore Saporito, Ingeborg Herold, Patrick Houthuizen, Hans van Assen, Hendrikus Korsten, Massimo Mischi
Abstract: Patients with acute heart failure often require urgent in-hospital treatment to treat symptoms, mainly due to congestion. There is urgent need for quantitative and minimally invasive techniques for the assessment of thoracic fluid status [1]. Indicator dilution measurements consist in the injection of an indicator whose concentration is sampled over time in the different cardiac chambers, yielding indicator dilution curves (IDCs). Physiological parameters, such as pulmonary blood volume and extravascular lung water, can then be estimated by fitting a model to the measured IDCs. Magnetic resonance imaging (MRI) contrast agents have been proposed as indicators, with the advantage of a non-invasive detection [2]. However, the definition of regions of interest (ROIs) in the dynamic contrast-enhanced MRI images is a time demanding process, usually performed manually. We propose a method to automatically define the ROIs and to extract the IDCs, based on statistical criteria rather than anatomical landmarks. Four-chamber views from 15 patients were obtained using an Intera scanner (Philips Healthcare) in combination with a cardiac coil array; bolus injections of Dotarem (Guerbet) were performed. A spoiled fast gradient echo sequence was used in combination with a saturation prepulse and parallel imaging. To identify different ROIs corresponding to different heart chambers, pixel based IDCs where clustered using a Gaussian mixture model. Dedicated preprocessing was used to isolate contrast enhancement from breathing artefacts. Clustering was performed in a subspace defined by the eigenvectors derived by principal component analysis of the considered IDCs. IDCs were fitted by the local density random walk (LDRW) model using a least-squares approach. Mean transit times (MTTs) for the left ventricle (LV) and right ventricle (RV) were estimated as well as their difference, the pulmonary transit time (PTT). The method was evaluated comparing the obtained IDCs’ kinetic parameters with those obtained from manually defined ROIs. The average difference on the MTT was 0.62±0.82 s and 0.37±0.87 s for the LV and RV MTT, respectively. For the PTT the difference was 0.25±0.92 s. When comparing the error to the average of the two the difference was 2.06±2.67 % and 2.22±4.81 % for the LV, and RV MTT, respectively, while it was 3.11±7.42 % for the PTT. A method for automatic ROI identification and IDC measurement is proposed. Estimated IDC kinetic parameters, such as MTT, show good agreement with those obtained manually.
15 mins
Salvatore Saporito, Patrick Houthuizen, Jean-Paul Aben, Hans van Assen, Massimo Mischi
Abstract: Left bundle branch block (LBBB) is a cardiac conduction anomaly resulting in dyssynchronous contraction of the left ventricle (LV). Cardiac resynchronization therapy (CRT) aims to restore the synchronous contractile activity of the heart. CRT has shown to be effective in reducing mortality and improving symptoms of heart failure patients with reduced ejection fraction and large electrocardiographic QRS complex duration [1]. However, up to one third of selected patients does not show signs of response to therapy. There is increasing evidence that LBBB patients are likely to respond to CRT. We propose a method based for the quantification of mechanical intraventricular synchronicity based on cine MRI. Short-axis multi-slice loops of the heart are acquired using a balanced steady-state free precession sequence. Automated segmentations of left ventricle endocardial and epicardial surfaces are obtained by CAAS MRV (Pie Medical Imaging BV). Time displacement curves representing radial wall motion are derived by tracking points on endocardial contours. After spatiotemporal regularization of the time displacement curves, regional contraction time (RCT) is determined by cross-correlation of the individual time displacement curves with a patient-specific reference. RCT can be visualized using polar bull’s eye plots. The method has been validated on 13 dogs before and after surgical induction of LBBB [2]. The standard deviation of the RCT distribution allows quantification of intraventricular dyssynchrony. Accuracy in identifying LBBB was compared to the commonly used standard deviation of time to peak contraction. LBBB dogs have larger standard deviation of RCT compared to controls (8.31 ± 2.63 % versus 3.30 ± 1.32 %, p<0.001), with a significant (p<0.001) increase after LBBB induction. Standard deviation of RCT shows better classification performances between the groups than standard deviation of time to peak (area under ROC curve of 0.99 and 0.86, with optimal threshold 6.1% and 9.1%, respectively). We have developed a method that can estimate regional contraction time from multi-slice cine MRI image series. Maps of regional dyssynchrony may be used to visualize contraction patterns to assist in CRT lead positioning. Further work will include validation against independent localized cardiac function measurements such as from tagged MRI and non-contact electrograms in a controlled animal study. A larger clinical validation in patients is also foreseen to assess the effectiveness of the proposed dyssynchrony measures for LBBB diagnosis and CRT response prediction. REFERENCES [1] Prinzen F. W. et al. ‘’Cardiac Resynchronization Therapy State-of-the-Art of Current Applications, Guidelines, Ongoing Trials, and Areas of Controversy’’, Circulation, 128(22), 2407-2418. (2013) [2] Vernooy K. et al. ‘’Cardiac resynchronization therapy cures dyssynchronopathy in canine left bundle-branch block hearts’, European Heart Journal, 28(17), 2148-2155. (2007)
15 mins
Hanne Kause, Olena Filatova, Remco Duits, Mark Bruurmijn, Andrea Fuster, Jos Westenberg, Luc Florack, Hans van Assen
Abstract: To improve the treatment outcome of cardiovascular disease, which is globally the leading cause of death (WHO 2014), it is important to develop methods for diagnosis and therapy assessment at early stages of the disease. It has been reported that persistent heart disease is preceded by changes in deformation and strain [1]. We propose a new method to obtain deformation and strain of the left ventricular wall from tagging MRI: Frequency Analysis for Deformation Estimation or FADE. The main principle behind FADE is the fact that spatial frequency of the tag pattern is directly linked to myocardial deformation, whereas it is unaffected by signal decay induced by T1 relaxation. Deformation and strain tensor fields can be retrieved from local frequency estimates given at least n (independent) tag observations, with n the spatial dimension. In previous studies [2,3], we tested FADE on synthetic, volunteer and patient tMRI data. We considered the conventional case of two tag directions, as well as the overdetermined case of four directions, which improves robustness. Additional scan time can be prevented by using one or two grid patterns consisting of multiple, simultaneously acquired tag directions, as is demonstrated on patient data. We tested FADE by comparing performance with respect to state-of-the-art method HARP [4] on synthetic data. Displacements were calculated for a large number of points organized in a rectangular grid and compared with ground truth. The average displacement errors with respect to their true counterparts are smaller for FADE than those obtained by HARP, 0.32±0.14 pixel versus 0.53±0.07 pixel. Strain results for volunteer and patient data are visually compared with corresponding linearized strain components derived from HARP. FADE does not require knowledge of material motion or tag line extraction. It exploits spatial tag frequency instead of amplitude information, and is therefore robust to tag fading. Moreover, it uses the full form strain tensor, instead of linearized strain. Additional testing is required to see if FADE significantly outperforms HARP on volunteer and patient data. Additionally, we would like to test different confidence measures for quality assessment.
15 mins
Hanne Kause, Patricia Márquez-Valle, Andrea Fuster, Aura Hernàndez-Sabaté, Luc Florack, Debora Gil, Hans van Assen
Abstract: Alterations in motion and deformation of the cardiac left ventricle are known to correlate with pathology [1]. Deformation can be extracted from cardiac tagging MRI (tMRI) using optical flow (OF) techniques, which use the fact that the harmonic phase of material points in a tMRI image is constant over time [2]. To apply this technique in a clinical setting, however, it is important to assess to what extent the performance of a particular OF method is stable across different clinical acquisition artifacts. In [3], we present a statistical validation framework to assess the motion and appearance factors that have the largest influence on OF accuracy. It consists of a 2-way ANOVA with factors given by OF methods (OFM) and the clinical sources of error (CSE), with the OF End-point-Error as ANOVA variable. The OFM factor groups are Harmonic Phase Flow (HPF), HPFC (HPF with constant weights) and HARP [4,5]. Acquisition, radiological noise and motion impacts constitute the CSE groups. This framework is applied to a database of synthetic tMRI images, featuring stripe and grid tags, which contains three motion patterns that are modelled with Rician noise and signal decay or free of noise without signal decay. The experiments show that HPF, which applies a regularity term only at areas where the phase is not reliable, performs better than HPFC, which uses a global regularity constraint. HARP has the worst performance, indicating the need for a regularity term to reduce the sensitivity to noise. Moreover, using a stripe or grid tag patterns does not affect the OF performance significantly, and motion assumptions do not bias the computations. Regarding the radiological artefacts, although there are no significant differences, we observed that OF performance is better for clean sequences. Therefore, we can conclude that the chosen OF methods are robust against CSE artifacts. This preliminary study encourages the use of the presented framework to explore OF performance in new settings. In the future we aim to apply it to clinical sequences to assess the impact of tMRI features on diagnostic quantities. Additionally, we will investigate the performance of different confidence measures (CM) on the optical flow methods used in this study.