Rizwan Ahmad - US grants

? DESCRIPTION (provided by applicant): Each year, more than 5 million Americans are diagnosed with cardiac valve disease. Among the valvular diseases, aortic valve stenosis (AVS) is the most common. With the aging population, the prevalence of AVS is expected to rise. Timing of intervention for AVS is largely based on the severity of stenosis and the presence of symptoms. Therefore, accurate assessment of the functional severity of stenosis is important when making clinical decisions regarding intervention. Currently, the most common non-invasive method for the assessment of AVS severity is Transthoracic Doppler echocardiography (TTE). However, many patients may have suboptimal evaluation with echocardiography due to poor acoustic windows, heavy calcification of the aortic valve, or significant flow acceleration in the left ventricular outflow tract which may obscure assessment of the aortic valve. For such patients, MRI-based 2D flow imaging (MRI-2DF), which is not impacted by acoustic windows, provides a viable alternative. MRI-2DF methods, however, are only sensitive to one directional component of the velocity vector; therefore, any misalignment of the velocity encoding direction with respect to the blood flow direction results in underestimation of the flow and velocity and, i turn, potential misclassification of disease severity. PC-MRI-based 4D flow imaging (MRI-4DF), with its volumetric spatial coverage and ability to encode all directions of the velocity, circumvents the shortcoming associated with TTE and MRI-2DF and thus can improve evaluation of AVS severity. The promise of MRI-4DF, however, is undone by prohibitively long scan times, which can be over 30 min. Despite recent efforts in utilizing parallel imaging, non-Cartesian trajectories, and compressive sensing (CS) inspired image recovery, MRI-4DF remains a research tool that is in need of further development to find clinical application. The goal of this work is to enable and demonstrate the feasibility of single breath-hold MRI-4DF in a small cohort of patients with AVS. In Specific Aim 1, we propose a novel technique, called Reconstructing Velocity Encoded MRI with Approximate message passing aLgorithms (ReVEAL), to reduce the acquisition time for MRI-4DF to a single breath-hold. In contrast to the existing CS techniques that utilize the underlying image sparsity, ReVEAL directly models the strong physical relationships inherent in the PC-MRI data. In particular, the proposed Bayesian approach capitalizes on the relationships in both magnitude and phase among the several velocity encodings. To solve the resulting Bayesian inference problem, an iterative image recovery method using message passing on a factor graph is proposed, yielding a fast algorithm with auto-tuning of all free parameters. In Specific Aim 2, we will use MRI-4DF data from a mechanical flow phantom and thirty AVS patients to validate the proposed approach. Preliminary results show that ReVEAL can accelerate MRI-2DF by a factor of 12; higher accelerations are expected for MRI-4DF due to added redundancy. This development will lead to more accurate characterization of cardiac valve disease than is possible with existing clinical methods.

Project Summary/Abstract Alterations in hemodynamics have been linked to wide-ranging cardiac and vascular conditions, including congenital heart disease, valvular abnormalities, aortic atherosclerosis and aneurysm, renal stenosis, portal hypertension due to liver cirrhosis, intracranial aneurysm and stenosis, and peripheral arterial disease. Phase- contrast MRI (PC-MRI) is a noninvasive imaging technique that can potentially provide a comprehensive evaluation of hemodynamics, which can be coupled with other important MRI-derived information on cardiovascular anatomy, function, and tissue characterization. However, the credibility of PC-MRI as a quantitative tool is challenged by the inaccuracies introduced by background phase. Studies have shown that this background phase can introduce significant errors in the quantification of flow. One method that has been proposed to quantify and correct for the background phase is to perform a separate scan using a static phantom. This method, despite being robust, is impractical because of the significant extra time required to perform phantom imaging for each clinical sequence performed. Another widely reported method to correct background phase is based on performing polynomial fitting to the pixels that belong to the static tissue. The accuracy of this method heavily relies on the availability of static tissue in the close vicinity of the region of interest?a requirement that is often not met when imaging the heart or great vessels. To address the issue of background phase that invariably impacts every PC-MRI measurement, we propose a new correction scheme called multi-slice acquisition and processing (mSAP). In mSAP, in addition to the slice of interest, at least one extra slice is collected using the same slice orientation and gradient waveforms but with a different table position. By jointly processing the background phase information from multiple slices, mSAP circumvents the shortcomings associated with existing methods at the cost of slightly prolonged acquisition. In Specific Aim 1, we will develop a data acquisition and processing method for mSAP. We will modify and streamline our current PC-MRI acquisition protocol to minimize the overhead associated with mSAP. To jointly process the multi-slice data, we will develop and implement polynomial regression based on generalized least squares with an ?1-norm penalty imposed on the coefficients of the polynomial. This fitting method is completely automated and does not require tuning parameters. In Specific Aim 2, we will validate mSAP using a pulsatile flow phantom and healthy volunteers. By using just one additional slice, we anticipate mSAP to reduce the background phase errors to the level where miscalculation of flow volume is reduced to below 5%. Our preliminary data demonstrate the validity of the primary assumption made in mSAP, i.e., background phase maps collected using the same gradient waveforms but different table positions are identical. We believe the methods developed in this work can be readily utilized in clinical settings to improve the accuracy of an otherwise potent imaging tool.

Heart failure, pulmonary hypertension, and congenital heart disease can compromise the capacity of the cardiovascular system to deliver sufficient oxygen (O2) to meet the varying metabolic demands of the organ systems. Mixed venous oxygen saturation measured in the main pulmonary artery is an accurate marker of the systemic delivery of blood and oxygen that provides key diagnostic and prognostic value. However, a true mixed venous saturation requires catheter access to the main pulmonary artery and may be underutilized as a diagnostic measure due to the associated level of invasiveness and risk. A non-invasive means to quickly and accurately measure O2 saturation (O2sat) in the main pulmonary artery, the cardiac chambers, and ideally anywhere in the body, would not only reduce the need for invasive catheterization procedures, but would also provide important physiological information that may be otherwise unavailable or unobtainable. In the blood, the magnetic resonance transverse relaxation time (T2) is related to the oxygen saturation of hemoglobin, and MR relaxometry has been previously proposed for in vivo estimation of blood O2 saturation; however, these estimates have relied on an impractical in vitro calibration on each patient, and results have been corrupted by flow-induced artifacts. A technique previously developed in our lab for rapid, single-shot T2 mapping has been modified to reduce flow artifacts and improve the accuracy of T2 measured in flowing blood. Together with this modified pulse sequence, we propose an entirely new approach to solving the Luz-Meiboom (L-M) equation that describes the relationship between T2 and O2sat in blood. We hypothesize that the use of varied preparation pulse timing along with direct measurement of easily accessible patient specific parameters will support the application of non-linear parameter estimation techniques to provide an accurate quantitative assessment of blood O2sat in the heart and deep vessels, even in locations having limited accessibility with other diagnostic techniques. We propose to optimize and validate this approach to non-invasive blood oximetry by meeting the following specific aims. Aim 1: We will define appropriate limits for acquisition parameters TE and 180 in a flow phantom and optimize acquisition parameters using statistical sensitivity analysis. Aim 2: We will empirically validate O2sat derived from T2 in a porcine model of graded hypoxemia that enables simultaneous acquisition of T2 and invasive O2sat measurement over a broad range of values. Aim 3: We will evaluate feasibility in a small cohort of heart failure patients undergoing clinically indicated pulmonary artery catheterization for mixed venous O2sat measurement. By addressing the flow sensitivity of the T2 preparation pulse and the inaccuracies introduced by oversimplification of the model relating T2 to O2sat, we anticipate that the level of accuracy and reproducibility for this technique will be raised to that required for clinical application in patients with cardiovascular disease. ?

Project Summary/Abstract Cardiovascular disease (CVD) claims more lives and costs more than any other diagnostic group in the USA. Cardiac magnetic resonance (CMR) is a non-invasive imaging tool that provides the most accurate and comprehensive assessment of the cardiovascular system, yet its role in clinical cardiology remains limited. A major impediment to wider usage of CMR is the inefficient acquisition that makes CMR exams excessively long, often lasting for more than an hour; this diminishes its efficiency and cost effectiveness relative to other modalities. The current paradigm offers either a prolonged segmented acquisition that requires regular cardiac rhythm and multiple breath-holds or a fallback option of real-time, free-breathing acquisition with degraded spatial and temporal resolutions that are below the Society for Cardiac Magnetic Resonance guidelines. The long-term goal of this investigation is to improve the diagnosis and evaluation of cardiovascular disease by transforming the existing segmented CMR acquisition into a more efficient protocol. The new paradigm will (i) eliminate the need to breath-hold, (ii) be effective in patients with arrhythmia, (iii) simplify the acquisition protocol, (iv) reduce the scan time, (v) provide whole-heart coverage, and (vi) enable spatial and temporal resolutions that rival the resolutions provided by segmented breath-held acquisition. In the last two decades, MRI technology has evolved rapidly. More recently, the combination of parallel MR imaging (pMRI) and compressive sensing (CS) recovery has been featured in numerous research studies and has delivered unprecedented acceleration. While pMRI has been adopted by the MRI industry and is available on almost all clinical platforms, CS recovery is still a long way away from routine clinical use. To bring CS recovery to clinical realm, there are a number of challenges that need to be addressed, including the well- recognized issues of long computation times and tuning parameters that require case-by-case adjustment. In this work, we will develop and validate a versatile CS recovery method, called sparsity adaptive composite recovery (SCoRe), that provides unmatched acceleration by exploiting sparsity across multiple representations. More importantly, SCoRe provides a data-driven tuning of all free parameters and thus eliminates the need to hand-tune regularization weights. Also, SCoRe is amenable to fast algorithms, and we expect the SCoRe-based image recovery to take only seconds on a GPU-based computing environment. We hypothesize that the proposed advances in data acquisition and processing will yield a new CMR protocol that is faster, easier for both patient and operator, and reliable over a broader spectrum of patients. We expect to achieve this objective by providing the necessary improvements in image quality (Aim 1), by reconstructing images in times suitable for clinical use (Aim 2), by validating the performance of the methods (Aim 3), and by demonstrating the effectiveness and efficiency of this new approach in a clinical trial (Aim 4).

Project Summary/Abstract Cardiac Magnetic Resonance (CMR) provides arguably the most comprehensive evaluation of the cardiovascular system; however, respiratory motion continues to adversely impact CMR, causing artifacts that lead to poor image quality, repeated scans, and decreased throughput, and thus represents a significant obstacle to clinical utility. For single-shot CMR, cardiac and breathing motions are ?frozen? by limiting the acquisition to an end-diastolic window less than 200 ms. For first pass perfusion, breathing motion cannot be eliminated because data from 50 to 60 consecutive heartbeats are required to capture contrast dynamics. For other single-shot applications such as late gadolinium enhancement (LGE) and parameter mapping, respiratory motion is introduced when the acquisition is repeated across several heartbeats to improve spatial and temporal resolution. To eliminate respiratory motion from single-shot images, non-rigid motion correction (MOCO) has been promoted as an attractive option that provides 100% acquisition efficiently. MOCO can be used either after the reconstruction or during the reconstruction. Such techniques, however, cannot account for through-plane motion, which can only be corrected prospectively, and can fail depending on image quality and the extent of motion. Prospective compensation of the respiratory motion has been recognized as an attractive alternative to existing gating and MOCO methods. Proposed methods use one or more navigator echoes?incompatible with or inefficient for many CMR protocols?to capture the respiratory motion and rely on simple parametric models that are inadequate to describe complex respiratory-induced cardiac motion. Due to these limitations, prospective methods have found limited applicability even in research settings. We propose a new framework to prospectively compensate respiratory motion. The proposed method, called PROspective Motion compensation using Pilot Tone (PROMPT), employs Pilot Tone technology and leverages machine learning principles to first learn complex respiratory-induced cardiac motion on a patient-specific basis and then prospectively compensate the motion by tracking the imaging plane, in real time, as a function of a Pilot Tone based respiratory signal. If successful, this synergistic combination of Pilot Tone and machine learning will lead to 100% efficiency for single-shot CMR exams performed under free-breathing conditions, will eliminate the need to setup navigator echoes, respiratory bellows, or other inefficient prospective gating measures, will minimize through-plane motion that can render the images non-diagnostic for CMR applications including fast-pass perfusion, parameter mapping, LGE, and coronary angiography, will provide a reliable surrogate measure of respiratory motion, and will facilitate highly accelerated compressive recovery.

PROJECT SUMMARY/ABSTRACT The primary goal of this investigation is to develop and validate a comprehensive, robust deep learning (DL) framework that improves MRI reconstruction beyond the limits of existing technology. The proposed framework uses ?plug-and-play? algorithms to combine physics-driven MR acquisition models with state-of-the-art learned image models, which are instantiated by image denoising subroutines. To fully exploit the rich structure of MR images, we propose to use DL-based denoisers that are trained in an application-speci?c manner. The proposed framework, termed PnP-DL, offers advantages over other existing DL methods, as well as compressed sensing (CS). Compared to existing DL methods for MRI reconstruction, PnP-DL is more immune to inevitable variations in the forward model, such as changes in the coil sensitivities or undersampling pattern, allowing it to generalize across applications and acquisition settings. Compared to CS, PnP-DL recovers images faster, with higher quality, and with potentially superior diagnostic value. Our preliminary results highlight the potential of PnP-DL to advance MRI technology. In this work, we will fur- ther develop PnP-DL and validate it in these major applications: cardiac cine, 2D brain, and 3D brain imaging. In Aim 1, we will train and optimize convolutional neural network-based application-speci?c denoisers for the above-mentioned applications. The denoiser with the best denoising performance will be selected for further investigation. In Aim 2, we will develop and compare different PnP algorithms. The algorithm yielding the best combination of reconstruction accuracy and computational speed will be implemented in Gadgetron for inline processing. In Aim 3, we will compare the performance of PnP-DL to other state-of-the-art methods using retro- spectively undersampled data. This study will demonstrate that, in terms of image quality, PnP-DL is superior to CS and existing DL methods and, despite higher acceleration, is non-inferior to parallel MRI with rate-2 acceler- ation. In Aim 4, we will evaluate the performance of PnP-DL using prospectively undersampled data from adult and pediatric patients. Successful completion of this project will demonstrate that PnP-DL outperforms state- of-the-art methods in terms of image quality while exhibiting a level of robustness and broad applicability that has eluded other DL-based MRI reconstruction methods. The acceleration and image quality improvement afforded by these developments will bene?t almost all MRI applications, including pediatric imaging, where reducing sedation is a pressing need, and high-dimensional imaging applications (e.g., whole-heart 4D ?ow imaging), which are too slow for routine clinical use.

Project Summary/Abstract Mitral valve regurgitation (MR) is a growing public health concern, and with an aging population, its prevalence is expected to rise steeply. For MR diagnosis and severity assessment, echocardiographic techniques have long been the standard of care. Assessment based on such techniques, however, has limitations, both in terms of technical challenges and treatment recommendations. As a result, optimal management of MR, especially determining the timing of surgery, remains complex and stands to benefit from tools that provide quantitative and comprehensive characterization of MR. The overall goal of this project is to develop and validate a stress cardiovascular MRI protocol that can lead to a more definitive treatment plan for MR patients. Cardiovascular MRI (CMR) is a well-established imaging technique that provides the most comprehensive evaluation of the cardiovascular system. The reproducibility of CMR-based flow quantification has been shown to be superior to that of echocardiography. Despite these advantages, the additive clinical value of CMR for MR patients has not been established. More recently, evidence has emerged that CMR-based assessment has better predictive power for clinical outcomes for MR patients and thus could play a central role in determining management plans for such patients. Existing CMR techniques, however, have significant limitations, precluding their use in routine clinical care. For example, the flow quantification using traditional 2D phase- contrast MRI (PC-MRI) is sensitive to the placement of the imaging plane, cannot measure the transvalvular flow directly, requires breath-holding, and is susceptible to irregular cardiac rhythm. Recently, 4D flow imaging, due to its volumetric coverage and three-directional encoding, has gained significant interest, but acquisition for 4D flow imaging using existing protocols can be prohibitively long, especially for whole-heart coverage. Also, existing 4D flow imaging protocols only perform imaging under resting conditions, which cannot fully characterize functional impairment that is only unmasked under stress testing. In this work, we will develop and validate a comprehensive CMR protocol that (i) provides ferumoxytol- enhanced 4D flow imaging with whole-heart coverage, (ii) requires minimal planning from the MRI technologist, (iii) is performed in clinically feasible acquisition time, (iv) does not require breath-holds or regular cardiac rhythm, (v) does not require navigator gating, (vi) allows imaging during exercise stress, exposing functional impairment, and (vii) additionally provides cardiac function quantification to explain and interpret stress-induced functional impairment observed in MR patients. In Aims 1 and 2, we will develop and optimize the protocol. In Aims 3 and 4, we will validate the accuracy and reproducibility of the protocol in 55 healthy subjects and 55 patients diagnosed with MR. We hypothesize that the developed protocol leads to a more reliable assessment of MR than possible with TTE alone and set the stage for larger clinical studies where the power of CMR parameters to predict clinical outcomes is demonstrated.