Purang Abolmaesumi

Heart failure (HF) is associated with poor patient outcomes and burdens healthcare systems and clinicians. Fortunately, therapeutic options are available for managing cardiac dysfunction if diagnosed early. Echocardiography (echo) can be used to assess cardiac function swiftly and detect signs or risk factors of HF. Nonetheless, echo acquisition and interpretation require extensive training and experience, leading to exceeding demand for the available clinical echo services. This thesis investigates the feasibility of machine learning (ML)-based solutions for analyzing heart function based on clinical echo data and available annotations. The goal is to automate measurements of indicators of functional diseases. We focus on guideline-aware supervised learning frameworks for assessing LV ejection fraction (EF), regional wall motion abnormality (WMA), and LV diastolic dysfunction (LVDD). We propose spatio-temporal neural networks to determine EF from echo cine loops. We utilize multi-task learning with observer variability modelling is leverage the label noise and decouple errors in different available EF labels. In the context of regional systolic function, we present an error quantification and visualization framework to evaluate the generalizability of disease-agnostic models on diseased cohorts. We validate segmentation models trained on standard populations in a WMA cohort and report global and local error metrics with weak wall segment labels. This framework enables us to further identify failure modes in trained ML models. Using the errors obtained from the weak labels, we observed that segmentation performance might become jeopardized in the presence of akinetic LV wall segments. Finally, in the most extensive study of its kind, we demonstrate the impacts of the updated clinical guidelines for diastolic function assessment based on measurements derived from echo. We propose a neural network to replicate the latest clinical guidelines for diastolic function classification and extend this model to a regression framework to obtain a novel continuous LVDD scoring system. Increasing the size and diversity of the training and test set for model training and clinical validation is critical to further developing ML-driven heart disease diagnostic tools. Future work may involve ML-based multi-chamber quantification, myocardium localization, and Doppler image analysis toward automatic disease diagnosis.

View record

Ultrasound-guided needle biopsy with pathologic grading is the standard-of-careto guide the systematic biopsy of the prostate. Systematic transrectal ultrasound isblind to prostate pathology; and diagnostic accuracy is still one of the main clinicalchallenges in prostate cancer treatment and management. Hence, there is a clearneed for improving US data and providing solutions to guide the biopsy procedure.Several methods including temporal enhanced UltraSound have been proposed toimprove US-based tissue typing. The ultimate clinical goal is to display the cancerlikelihood maps on B-mode ultrasound images in real time to help with indicationof cancer in the core, decision support for biopsy targeting, and eventually reduc-ing the number of unnecessary biopsies. The objective of this dissertation is todeploy this by integration of temporal enhanced UltraSound and machine learningapproaches. Towards fulfilling this objective, in this dissertation a weakly super-vised learning technique is utilized to learn from ultrasound image regions associ-ated with the corresponding data on patient level. To improve the prostate cancerdetection we consider the nonstationarity nature of the data and use a complexneural network to find a better representation of the data in the embedding spacefor a better classification. Later, we automate reliable detection by estimatingthe model and label uncertainty. We finally show in order to improve the performanceof prostate cancer classification, the label noise needs to be considered and in thiswork, the latter is done by implementing a label refinement technique.

View record

Echocardiography (echo) plays an important role in cardiac imaging and provides a non-invasive, low-cost, and widely available diagnostic tool for the comprehensive evaluation of cardiac structure and function. However, ultrasound image interpretation remains a challenge, particularly for the novices. In this thesis, I propose several machine learning methods with the aim of helping in cardiac assessment. In particular, the proposed methods take advantage of novel advancements in spatio-temporal analysis of videos to find a better representation of the cardiac cine series.First, I propose a customized supervised learning method in order to find two specific phases in a cardiac cycle. Specifically, identification of the end-systolic (ES) and end-diastolic (ED) phases from the echo cine series is a critical step in the quantification of cardiac chamber size and function. Later, I develop a self-supervised learning method for frame-rate up-conversion to augment conventional imaging without the need of specialized beamforming and imaging hardware. In another work, I propose a self-supervised learning framework to synchronize various cross-sectional 2D echo videos without any human supervision or external inputs. I show that such rich, yet free semantic representation can be used not only for synchronization of multiple cine series, but also for fine-grained cardiac phase detection. Finally, I propose a semi-supervised learning method to detect the cardiac rhythm based solely on echo without the need for an electrocardiogram (ECG), which is commonly used for cardiac rhythm detection.

View record

The continuous advent of novel imaging technologies in the past two decades has created new avenues for biomechanical modeling, biomedical image analysis, and machine learning. While there still is relatively a long way ahead of the biomedical tools for them to be integrated into the conventional clinical practice, biomechanical modeling and machine learning have shown noticeable potential to change the future of treatment planning. In this work, we focus on some of the challenges in the modeling of the masticatory (chewing) system for the treatment planning of jaw reconstructive surgeries. Here, we discuss novel methods to capture the kinematics of the human jaw, fuse information in between imaging modalities, estimate the missing parts of the 3D structures (bones), and solve the inverse dynamics problem to estimate the muscular forces. This research is centered around the human masticatory system and its core component, the mandible (jaw), while focusing on the treatment planning for cancer patients. We investigate jaw tracking and develop an optical tracking system using subject-specific dental attachments and infrared markers. To achieve that, a fiducial localization method was developed to increase the accuracy of tracking. In data fusion, we propose a method to register the 3D dental meshes on the MRI of the maxillofacial structures. We use fatty ellipsoidal objects, which resonate in MRI, as fiducial landmarks to automate the entire workflow of data fusion. In shape completion, we investigate the feasibility of generating a 3D anatomy from a given dense representation using deep neural architectures. We then extend on our deep method to train a probabilistic shape completion model, which takes a variational approach to fill in the missing pieces of a given anatomy. Lastly, we tackle the challenge of inverse dynamics and motor control for biomechanical systems where we investigate the applicability of reinforcement learning (RL) for muscular force estimation. With the mentioned portfolio of methods, we try to make biomechanical modeling more accessible for clinicians, either via automating known manual processes or introducing new perspectives.

View record

Heart disease is one of the foremost causes of mortality worldwide. Echocardiography (echo) is a commonly used modality to study the heart's structure and function as it is non-invasive and cost-effective. In recent years, the emergence of readily accessible ultrasound (US) devices, namely point-of-care ultrasound (POCUS), has accelerated the widespread uptake of echo. However, robust echo acquisition and interpretation is tedious, and its efficacy depends on the skills of expert physicians. In this thesis, we investigate machine learning solutions to promote reliable computer-assisted echo examination, improving interpretation accuracy, diagnostic throughput, and test-retest reliability. The main tackled challenges include considering temporal data dependencies, label sparsity, multi-task learning, low-quality noisy nature of echo, and noisy clinical labels with high inter- and intra-observer variability. We present a deep spatio-temporal model integrating recurrent fully convolutional neural networks and optical flow estimation maps to accurately track the left ventricle in echo video clips. We also propose a semi-supervised learning algorithm to leverage unlabeled data to improve the performance of machine learning methods. Moreover, we present a computationally efficient mobile framework for accurate left ventricular ejection fraction estimation. The proposed mobile application runs in real time on an Android smartphone with a connection to a POCUS or cart-based ultrasound device. We also suggest adapting conditional generative adversarial network (cGAN) architectures to improve the quality of echo data. Further, we investigate predictive uncertainties via Bayesian deep learning to sustain the robust deployment of the developed methodologies.

View record

Prostate cancer is the second most prevalent cancer in men worldwide. Magnetic Resonance Imaging (MRI) is widely used for prostate cancer diagnosis and guiding biopsy procedures due to its ability in providing superior contrast between cancer and adjacent soft tissue. Appropriate clinical management of prostate cancer critically depends on meticulous detection and characterization of the disease and precise biopsy procedures if necessary. The goal of this thesis is to develop computational methods to aid radiologists in diagnosing prostate cancer in MRI and planning necessary interventions. To this end, we have developed novel methods for assessing probability of clinically significant prostate cancer in MRI, localizing biopsy needles in MRI, and providing segmentation of structures such as the prostate gland.The proposed methods in this thesis are based on supervised machine learning techniques, in particular deep convolutional neural networks (CNNs). We have also developed methodology that is necessary in order for such deep networks to eventually be useful in clinical decision-making workflows; this spans the areas of domain adaptation, confidence calibration, and uncertainty estimation for CNNs. We used domain adaptation to transfer the knowledge of lesion segmentation learned from MRI images obtained using one set of acquisition parameters to another. We also studied predictive uncertainty in the context of medical image segmentation to provide model confidence (i.e expectation of success) at inference time. We further proposed parameter ensembling by perturbation for calibration of neural networks.

View record

Biomechanical modelling has the potential to start the next revolution in medicine, just as imaging has done in decades past. Current technology can now capture extremely detailed information about the structure of the human body. The next step is to consider function. Unfortunately, though there have been recent advances in creating useful anatomical models, there are still significant barriers preventing their widespread use.In this work, we aim to address some of the major challenges in biomechanical model construction. We examine issues of discretization: methods for representing complex soft tissue structures; issues related to consolidation of data: how to register information from multiple sources, particularly when some aspects are unreliable; and issues of detail: how to incorporate information necessary for reproducing function while balancing computational efficiency.To tackle discretization, we develop a novel hex-dominant meshing approach that allows for quality control. Our pattern-base tetrahedral recombination algorithm is extremely simple, and has tight computational bounds. We also compare a set of non-traditional alternatives in the context of muscle simulation to determine when each might be appropriate for a given application.For the fusion of data, we introduce a dynamics-driven registration technique which is robust to noise and unreliable information. It allows us to encode both physical and statistical priors, which we show can reduce error compared to the existing methods. We apply this to image registration for prostate interventions, where only parts of the anatomy are visible in images, as well as in creating a subject-specific model of the arm, where we need to adjust for both changes in shape and in pose.Finally, we examine the importance of and methods to include architectural details in a model, such as muscle fibre distribution, the stiffness of thin tendinous structures, and missing surface information. We examine the simulation of muscle contractions in the forearm, force transmission in the masseter, and dynamic motion in the upper airway to support swallowing and speech simulations.By overcoming some of these obstacles in biomechanical modelling, we hope to make it more accessible and practical for both research and clinical use.

View record

The ultimate diagnosis of prostate cancer involves histopathology analysis of tissue samples obtained through prostate biopsy, guided by either transrectal ultrasound (TRUS), or fusion of TRUS with multi-parametric magnetic resonance imaging. Appropriate clinical management of prostate cancer requires accurate detection and assessment of the grade of the disease and its extent. Despite recent advancements in prostate cancer diagnosis, accurate characterization of aggressive lesions from indolent ones is an open problem and requires refinement.Temporal Enhanced Ultrasound (TeUS) has been proposed as a new paradigm for tissue characterization. TeUS involves analysis of a sequence of ultrasound radio frequency (RF) or Brightness (B)-mode data using a machine learning approach. The overarching objective of this dissertation is to improve the accuracy of detecting prostate cancer, specifically the aggressive forms of the disease and to develop a TeUS-augmented prostate biopsy system. Towards full-filling this objective, this dissertation makes the following contributions: 1) Several machine learning techniques are developed and evaluated to automatically analyze the spectral and temporal aspect of backscattered ultrasound signals from the prostate tissue, and to detect the presence of cancer; 2) a patient-specific biopsy targeting approach is proposed that displays near real-time cancer likelihood maps on B-mode ultrasound images augmenting their information; and 3) the latent representations of TeUS, as learned by the proposed machine learning models, are investigated to derive insights about tissue dependent features residing in TeUS and their physical interpretation.A data set consisting of biopsy targets in mp-MRI-TRUS fusion-biopsies with 255 biopsy cores from 157 subjects was used to generate and evaluate the proposed techniques. Clinical histopathology of the biopsy cores was used as the gold-standard. Results demonstrated that TeUS is effective in differentiating aggressive prostate from clinically less-significant disease and non-cancerous tissue. Evidence derived from simulation and latent-feature visualization showed that micro-vibrations of tissue microstructure, captured by low-frequency spectral features of TeUS, is a main source of tissue-specific information that can be used for detection of prostate cancer.

View record

Visualizing vertebrae or other bone structures clearly in ultrasound imaging is important for many clinical applications such as ultrasound-guided spinal needle injections and scoliosis detection. Another growing research topic is fusing ultrasound with other imaging modalities to get the benefit from each modality. In such approaches, tissue with strong interfaces, such as bones, are typically extracted and used as the feature for registration. Among those applications, the spine is of particular interest in this thesis. Although such ultrasound applications are promising, clear visualization of spine structures in ultrasound imaging is difficult due to factors such as specular reflection, off-axis energy and reverberation artifacts. The received channel ultrasound data from the spine are often tilted even after delay correction, resulting in signal cancellation during the beamforming process. Conventional beamformers are not designed to tackle this issue. In this thesis, we propose three beamforming methods dedicated to improve the visualization of spine structures. These methods include an adaptive beamforming method which utilizes the accumulated phase change across the receive aperture as the beamforming weight. Then, we propose a log-Gabor based directional filtering method to regulate the tilted channel data back to the beamforming direction to avoid bone signal cancellation. Finally, we present a closed-loop beamforming method which feeds back the location of the spine to the beamforming process so that backscattered bone signals can be aligned prior-to the beamforming. Field II simulation, phantom and in vivo results confirm significant contrast improvement of spinal structures compared with the conventional delay-and-sum beamforming and other adaptive beamforming methods.

View record

Scaphoid fracture is the most probable outcome of wrist injury and it often occurs due to sudden fall on an outstretched arm. To fix an acute non-displaced fracture, a volar percutaneous surgical procedure is highly recommended as it provides faster healing and better biomechanical outcome to the recovered wrist. Conventionally, this surgical procedure is performed under X-ray based fluoroscopic guidance, where surgeons need to mentally determine a trajectory of the drilling path based on a series of 2D projection images. In addition to challenges associated with mapping 2D information to a 3D space, the process involves exposure to ionizing radiation. Ultrasound (US) has been suggested as an alternate; US has many advantages including its non-ionizing nature and real-time 3D acquisition capability. US images are, however, difficult to interpret as they are often corrupted by significant amounts of noise or artifact, in addition, the appearance of the bone surfaces in an US image contains only a limited view of the true surfaces. In this thesis, I propose techniques to enable ultrasound guidance in scaphoid fracture fixation by augmenting intraoperative US images with preoperative computed tomography (CT) images via a statistical anatomical model of the wrist. One of the major contributions is the development of a multi-object statistical wrist shape+scale+pose model from a group of subjects at wide range of wrist positions. The developed model is then used to register with the preoperative CT to obtain the shapes and sizes of the wrist bones. The intraoperative procedure starts with a novel US bone enhancement technique that takes advantage of an adaptive wavelet filter bank to accurately highlight the bone responses in US. The improved bone enhancement in turn enables a registration of the statistical pose model to intraoperative US to estimate the optimal scaphoid screw axis for guiding the surgical procedure. In addition to this sequential registration technique, I propose a joint registration technique that allows a simultaneous fusion of the US and CT data for an improved registration output. We conduct a cadaver experiment to determine the accuracy of the registration process, and compare the results with the ground truth.

View record

Low-dose-rate prostate brachytherapy is a minimally invasive treatment approach for localized prostate cancer. It takes place in one session by permanent implantation of several small radio-active seeds inside and adjacent to the prostate. The current procedure at the majority of institutions requires planning of seed locations prior to implantation from transrectal ultrasound (TRUS) images acquired weeks in advance. The planning is based on a set of contours representing the clinical target volume (CTV). Seeds are manually placed with respect to a planning target volume (PTV), which is an anisotropic dilation of the CTV, followed by dosimetry analysis. The main objective of the plan is to meet clinical guidelines in terms of recommended dosimetry by covering the entire PTV with the placement of seeds. The current planning process is manual, hence highly subjective, and can potentially contribute to the rate and type of treatment related morbidity. The goal of this thesis is to reduce subjectivity in prostate brachytherapy planning. To this end, we developed and evaluated several frameworks to automate various components of the current prostate brachytherapy planning process. This involved development of techniques with which target volume labels can be automatically delineated from TRUS images. A seed arrangement planning approach was developed by distributing seeds with respect to priors and optimizing the arrangement according to the clinical guidelines. The design of the proposed frameworks involved the introduction and assessment of data fusion techniques that aim to extract joint information in retrospective clinical plans, containing the TRUS volume, the CTV, the PTV and the seed arrangement. We evaluated the proposed techniques using data obtained in a cohort of 590 brachytherapy treatment cases from the Vancouver Cancer Centre, and compare the automation results with the clinical gold-standards and previously delivered plans. Our results demonstrate that data fusion techniques have the potential to enable automatic planning of prostate brachytherapy.

View record

Prostate biopsy is the gold standard for cancer diagnosis. This procedure is guided using a 2D transrectal ultrasound (TRUS) probe. Unfortunately, early stage tumors are not visible in ultrasound and prostate motion/deformations make targeting challenging. This results in a high number of false negatives and patients are often required to repeat the procedure. Fusion of magnetic resonance images (MRI) into the workspace of a prostate biopsy has the potential to detect tumors invisible in TRUS. This allows the radiologist to better target early stage cancerous lesions. However, due to different body positions and imaging settings, the prostate undergoes motion and deformation between the biopsy coordinate system and the MRI. Furthermore, due to variable probe pressure, the prostate moves and deforms during biopsy as well. This introduces additional targeting errors. A biopsy system that compensates for these sources of error has the potential to improve the targeting accuracy and maintain a 3D record of biopsy locations. The goal of this thesis is to provide the necessary tools to perform freehand MR-TRUS fusion for prostate biopsy using a 3D guidance system. To this end, we have developed two novel surface-based registration methods for incorporating the MRI into the biopsy workspace. The proposed methods are the first methods that are robust to missing surface regions for MR-TRUS fusion (up to 30% missing surface points). We have validated these fusion techniques on 19 biopsy, 10 prostatectomy and 11 brachytherapy patients. In this thesis, we have also developed methods that combine intensitybased information with biomechanical constraints to compensate for prostate motion and deformations during the biopsy. To this end, we have developed a novel 2D-3D registration framework, which was validated on an additional 10 biopsy patients. Our results suggest that accurate 2D-3D registration for freehand biopsy is feasible.The results presented suggest that accurate registration of MR and TRUS data in the presence of partially missing data is feasible. Moreover, we demonstrate that in the presence of variable probe pressure during freehand biopsy, a combination of intensity-based and biomechanically constrained 2D-3D registration can enable accurate alignment of pre-procedure TRUS with 2D real time TRUS images.

View record

Diagnosis and clinical management of neurological disorders that affect brain structure, function and networks would benefit substantially from the development of techniques that combine multi-modal and/or multi-task information. Here, we propose a joint Source Based Analysis (jSBA) framework to identify common information across structural and functional contrasts in data from MRI and fMRI experiments, for classification of individuals with neurological and psychiatric disorders. The framework consists of three components: 1) individual's feature generation, 2) joint group analysis, and 3) classification of individuals based on the group's generated features. In the proposed framework, information from brain neuroimaging datasets is reduced to a feature that is a lower-dimensional representation of a selected brain structure or task-related activation pattern. For each individual, features are used within a joint analysis method to generate basis brain activation sources and their corresponding modulation profiles. Modulation profiles are used to classify individuals into different categories. We perform two experiments to demonstrate the potential of the proposed framework to classify groups of subjects based on structural and functional brain data. In the fMRI analysis, functional contrast images derived from a study of auditory and speech perception of 16 young and 16 older adults are used for classification of individuals. First, we investigate the effect of using multi-task fMRI data to improve the classification accuracy. Then, we propose a novel joint Sparse Representation Analysis (jSRA) to identify common information across different functional contrasts in data. We further assess the reliability of jSRA, and visualize the brain patterns obtained from such analysis. In the sMRI analysis, features representing position, orientation and size (i.e. pose), shape, and local tissue composition of brain are used to classify 19 depressed and 26 healthy individuals. First, we incorporate pose and shape measures of morphology, which are not usually analyzed in neuromorphometric studies, to measure structural changes. Then, we combine brain tissue composition and morphometry using the proposed jSBA framework. In a cross-validation leave-one-out experiment, we show that we can classify the subjects with an accuracy of 67% solely based on the information gathered from the joint analysis of features obtained from multiple brain structures.

View record

Ultrasound is a safe, portable, inexpensive and real-time modality that can produce 2D and 3D images. It is a valuable intra-operative imaging modality to guide surgeons aiming to achieve higher accuracy of the intervention and improve patient outcomes. In all the clinical applications that use tracked ultrasound, one main challenge is to precisely locate the ultrasound image pixels with respect to a tracking sensor on the transducer. This process is called spatial calibration and the objective is to determine the spatial transformation between the ultrasound image coordinates and a coordinate system defined by the tracking sensor on the transducer housing. Another issue in ultrasound guided interventions is that tracking surgical tools (for example an epidural needle) usually requires expensive, large optical trackers or low accuracy magnetic trackers and there is a need for a low-cost, easy-to-use and accurate solution. In this thesis, for the first problem I have proposed two novel complementary methods for ultrasound calibration that provide ease of use and high accuracy. These methods are based on my differential technique which enables high measurement accuracy. I developed a closed-form formulation that makes it possible to achieve high accuracy with using a low number of images. For the second problem, I developed a method to track surgical tools (epidural needles in particular) using a single camera mounted on the ultrasound transducer to facilitate ultrasound guided interventions. The first proposed ultrasound calibration method achieved an accuracy of 0.09 ± 0.39 mm. The second method with a much simpler phantom yet achieved similar accuracy compared to the N-wire method. The proposed needle tracking method showed high accuracy of 0.94 ± 0.46 mm.

View record

The idea of full six degree-of-freedom tracking of ultrasound images solely based on speckle information has been a long term research goal. It would eliminate the need for any additional tracking hardware and reduces cost and complexity of ultrasound imaging system, while providing the benefits of three-dimensional imaging.Despite its significant promise, speckle tracking has proven challenging due to several reasons including the dependency on a rare kind of speckle pattern in real tissue, underestimation in the presence of coherency or specular reflection, ultrasound beam profile spatial variations, need for RF (Radio Frequency) data, and artifacts produced by out-of-plane rotation. So, there is a need to improve the utility of freehand ultrasound in clinics by developing techniques to tackle these challenges and evaluate the applicability of the proposed methods for clinical use.We introduce a model-fitting method of speckle tracking based on the Rician Inverse Gaussian (RiIG) distribution. We derive a closed-form solution of the correlation coefficient of such a model, necessary for speckle tracking. In this manner, it is possible to separate the effect of the coherent and the non-coherent part of each patch. We show that this will increase the accuracy of the out-of-plane motion estimation.We also propose a regression-based model to compensate for the spatial changes of the beam profile.Although RiIG model fitting increases the accuracy, it is only applicable on ultrasound sampled RF data and computationally expensive. We propose a new framework to extract speckle/noise directly from B-mode images and perform speckle tracking on the extracted noise. To this end, we investigate and develop Non-Local Means (NLM) denoising algorithm based on a prior noise formation model.Finally, in order to increase the accuracy of the 6-DoF transform estimation, we propose a new iterative NLM denoising filter for the previously introduced RiIG model based on a new NLM similarity measure definition. The local estimation of the displacements are aggregated using Stein’s Unbiased Risk Estimate (SURE) over the entire image. The proposed filter-based speckle tracking algorithm has been evaluated in a set of ex vivo and in vivo experiments.

View record

The blind placement of an epidural needle is among the most difficult regional anesthetic techniques. The challenge is to insert the needle in the midline plane of the spine and to avoid overshooting the needle into the spinal cord. Prepuncture 2D ultrasound scanning has been introduced as a reliable tool to localize the target and facilitate epidural needle placement. Ideally, real-time ultrasound should be used during needle insertion to monitor the progress of needle towards the target epidural space. However, several issues inhibit the use of standard 2D ultrasound, including the obstruction of the puncture site by the ultrasound probe, low visibility of the target in ultrasound images of the midline plane, and increased pain due to a longer needle trajectory. An alternative is to use 3D ultrasound imaging, where the needle and target could be visible within the same reslice of a 3D volume; however, novice ultrasound users (i.e., many anesthesiologists) have difficulty interpreting ultrasound images of the spine and identifying the target epidural space. In this thesis, I propose techniques that are utilized for augmentation of 3D ultrasound images with a model of the vertebral column. Such models can be pre-operatively generated by extracting the vertebrae from various imaging modalities such as Computed Tomography (CT) or Magnetic Resonance Imaging (MRI). However, these images may not be obtainable (such as in obstetrics), or involve ionizing radiation. Hence, the use of Statistical Shape Models (SSM) of the vertebrae is a reasonable alternative to pre-operative images. My techniques include construction of a statistical model of vertebrae and its registration to ultrasound images. The model is validated against CT images of 56 patients by evaluating the registration accuracy. The feasibility of the model is also demonstrated via registration to 64 in vivo ultrasound volumes.

View record