Mrudang Mathur

All Publications

• A Multitask Deep Learning Model for Pediatric Echocardiography Analysis. Circulation Cho, J., Mathur, M., Kaur, D., Duda, M., Dahlan, A., Krishnan, A., Leipzig, M., Shad, R., Gonzalez, A. K., Logan, J., Seidman, C., Fong, R., Kumar, A., Zakka, C., Carter, E., Padiyath, A., Jones, A., Quartermain, M. D., Langlotz, C. P., Jolley, M. A., Hiesinger, W. 2026

  Abstract

  Congenital heart defects afflict ≈1% of all births worldwide. Although deep learning has shown significant promise in automating and improving adult echocardiography analysis, existing pediatric-based models are often limited to single tasks and specific echocardiographic views. To address this, we introduce EchoAI-Peds, a multitask deep learning model for pediatric echocardiography. Our model was developed using the most comprehensive set of pediatric labels to date and is designed to integrate information from multiple views simultaneously.We trained a video-based vision transformer to simultaneously detect 28 congenital heart defects, structural and functional abnormalities, repairs, and interventions directly from complete pediatric echocardiography studies with multiple videos. Our model was developed using >700 000 videos derived from >12 000 studies performed at Stanford Medicine between 2014 and 2021. Specifically, our model was trained on 10 815 studies (median age, 8 [IQR 2-14] years; 45% girls) and validated on 1294 studies (median age, 9 [IQR 2-15] years; 46% girls). Model efficacy was tested on an internal held-out data set of 1336 studies from that same period. In addition, model generalizability was tested on a spatially and temporally distinct patient cohort at the Children's Hospital of Philadelphia using 2121 studies performed between 2024 and 2025.Our model achieved macroaveraged (area under the receiver operating characteristic curve (AUROC) values of 0.91 (95% CI, 0.90-0.92) on the internal test set (median age, 8 [IQR 2-14] years; 47% girls) and AUROC values of 0.89 (95% CI, 0.88-0.90) on the external test set (median age 6 [IQR 0.92-13] years; 44.4% girls). Moreover, EchoAI-Peds significantly outperformed adult-based echocardiography foundation models trained on substantially larger data sets, including EchoCLIP (internal AUROC=0.58, 95% CI, 0.56-0.60; external AUROC=0.61, 95% CI, 0.59-0.62) and EchoPrime (internal AUROC=0.58, 95% CI, 0.57-0.61; external AUROC=0.60, 95% CI, 0.59-0.62). Finally, our model demonstrated robust performance across patient age, patient sex, and studies with varying number of videos.Our findings demonstrate the remarkable potential for multitask deep learning models to aid the interpretation of pediatric echocardiograms. In addition, our results underscore the need for models that are specifically tailored to pediatric populations.

  View details for DOI 10.1161/CIRCULATIONAHA.126.080619

  View details for PubMedID 42517220

• Understanding the mechanisms behind the annuloplasty effect in tricuspid valve TEER: a computational study. Journal of biomechanics Haese, C. E., Dubey, V., Mathur, M., Kreidel, F., Fuhg, J. N., Moussa, I. D., Timek, T. A., Rausch, M. K. 2026; 205: 113423

  Abstract

  An annuloplasty effect has been observed following tricuspid transcatheter edge-to-edge repair (TEER) and is associated with a therapeutic benefit. However, the mechanisms underlying the annuloplasty effect remain unknown. In this study, we investigated the impact of TEER-induced annular forces on the annuloplasty effect. Additionally, we explored the influence of clip size, clip orientation, leaflet pair, and leaflet site on TEER-induced annular forces. To this end, we simulated 34 TEER repairs in finite element models of three human tricuspid valves. We used an NTW or XTW TriClip, placed between either the anterior-septal or anterior-posterior leaflet pairs at a central or near-annulus site. For each scenario, we quantified the reduction in annular area, septal-lateral (SL) diameter, and anterior-posterior (AP) diameter. We also reported the total annular force, orientation of maximum annular force, total papillary muscle force, leaflet stress, and coaptation area ratio following TEER. We found that TEER induced annular forces (2.00 ± 1.68 N), which were associated with the annuloplasty effect as measured by reduction in annular area (3.11±2.29%, p < 0.0001) and SL diameter (2.10±2.25%, p < 0.0001). The maximum annular force aligned with the orientation of the clip (p < 0.0001). Furthermore, larger XTW clips induced more annular force (p < 0.0001) and leaflet stress (p < 0.0001) than smaller NTW clips. A central anterior-posterior site induced more force than a near-annulus anterior-posterior site (p = 0.0166), but showed no difference from an anterior-septal pair (p = 0.939). In summary, we demonstrated that TEER procedural parameters alter the magnitude of induced annular forces, which, in turn, correlate with the degree of annuloplasty effect.

  View details for DOI 10.1016/j.jbiomech.2026.113423

  View details for PubMedID 42364443

• A generalizable deep learning system for cardiac MRI. Nature biomedical engineering Shad, R., Zakka, C., Kaur, D., Mathur, M., Fong, R., Cho, J., Filice, R. W., Mongan, J., Kallianos, K., Khandwala, N., Eng, D., Leipzig, M., Witschey, W. R., de Feria, A., Ferrari, V. A., Ashley, E. A., Acker, M. A., Langlotz, C., Hiesinger, W. 2026

  Abstract

  Cardiac MRI allows for a comprehensive assessment of myocardial structure, function and tissue characteristics. Here we describe a foundational vision system for cardiac MRI, capable of representing the breadth of human cardiovascular disease and health. Our deep-learning model is trained via self-supervised contrastive learning, in which visual concepts in cine-sequence cardiac MRI scans are learned from the raw text of the accompanying radiology reports. We train and evaluate our model on data from four large academic clinical institutions in the United States. We additionally showcase the performance of our models on the UK BioBank and two additional publicly available external datasets. We explore emergent capabilities of our system and demonstrate remarkable performance across a range of tasks, including the problem of left-ventricular ejection fraction regression and the diagnosis of 39 different conditions such as cardiac amyloidosis and hypertrophic cardiomyopathy. We show that our deep-learning system is capable of not only contextualizing the staggering complexity of human cardiovascular disease but can be directed towards clinical problems of interest, yielding impressive, clinical-grade diagnostic accuracy with a fraction of the training data typically required for such tasks.

  View details for DOI 10.1038/s41551-026-01637-3

  View details for PubMedID 41882174

• Sex disparities in deep learning estimation of ejection fraction from cardiac magnetic resonance imaging. NPJ digital medicine Kaur, D., Shad, R., Kumar, A., Mathur, M., Cho, J., Fong, R., Zakka, C., Phillips, C., Hiesinger, W. 2026

  Abstract

  The advent of artificial intelligence in cardiovascular imaging holds immense potential for earlier diagnoses, precision medicine, and improved disease management. However, the presence of sex-based disparities and strategies to mitigate biases in deep learning models for cardiac imaging remain understudied. In this study, we analyzed algorithmic bias in a foundation model that was pretrained on cardiac magnetic resonance imaging and radiology reports from multiple institutes and finetuned to estimate ejection fraction (EF) on the UK Biobank dataset. The model performed significantly worse in EF estimation for females than males in the diagnosis of reduced EF. Algorithmic fairness did not improve despite masking of protected attributes in radiology reports and data resampling, although explicit input of sex in model finetuning may improve EF estimation in some cases. The underdiagnosis of reduced EF among females holds critical implications for the exacerbation of existing sex-based disparities in cardiovascular health. We advise caution in the development of models for cardiovascular imaging to avoid such pitfalls.

  View details for DOI 10.1038/s41746-025-02330-6

  View details for PubMedID 41577988

• Tricuspid valve leaflet remodeling in sheep with biventricular heart failure: A comparison between leaflets. Acta biomaterialia Kostelnik, C. J., Meador, W. D., Lin, C., Mathur, M., Malinowski, M., Jazwiec, T., Malinowska, Z., Piekarska, M. L., Gaweda, B., Timek, T. A., Rausch, M. K. 2025

  Abstract

  Tricuspid valve leaflets are dynamic tissues that can respond to altered biomechanical and hemodynamic loads. Each leaflet has unique structural and mechanical properties, leading to differential in vivo strains. We hypothesized that these intrinsic differences drive heterogeneous, disease-induced remodeling between the leaflets. Although we previously reported significant remodeling changes in the anterior leaflet, the responses among the other two leaflets have not been reported. Using a sheep model of biventricular heart failure, we compared the remodeling responses between all tricuspid leaflets. Our results show that the anterior leaflet underwent the most significant remodeling, while the septal and posterior leaflets exhibited similar but less pronounced changes. We found several between-leaflet differences in key structural and mechanical metrics that have been shown to contribute to valvular dysfunction. Diseased animals exhibited significantly larger septal and anterior leaflets, thicker anterior and posterior leaflets, and stiffer anterior leaflets. Additionally, the septal leaflet's anisotropy index significantly decreased. Also, the septal and anterior leaflets showed increased collagen fiber dispersion near the atrial surface. As for remodeling markers, alpha-smooth muscle actin (alpha-SMA), Ki67, matrix-metalloprotease 13 (MMP13), and transforming growth factor beta-1 (TGF-beta1) were upregulated in spatially and leaflet-dependent patterns. That is, we observed increased expression of these markers within septal leaflets' near-annulus and belly regions, increased expression in anterior leaflets' belly region, and varied expression in posterior leaflets. These findings underscore the need to consider leaflet-specific remodeling to fully understand tricuspid valve dysfunction and to develop targeted therapies for its treatment and more accurate computational models. STATEMENT OF SIGNIFICANCE: Our study is significant as it advances our understanding of tricuspid valve remodeling by providing a comprehensive analysis of all three leaflets in a sheep model of biventricular heart failure. Unlike prior works that focused primarily on the anterior leaflet or generalized leaflet changes, we integrated morphological, histological, immunohistochemistry, biaxial mechanical testing, and two-photon microscopy to quantify differences between all three tricuspid valve leaflets (anterior, posterior, and septal) across multiple functional scales. This comprehensive approach highlights the unique remodeling response of each leaflet. Our findings offer critical insights for developing targeted therapeutic strategies and improving computational models of disease progression.

  View details for DOI 10.1016/j.actbio.2025.03.052

  View details for PubMedID 40180007

• Tricuspid valve edge-to-edge repair simulations are highly sensitive to annular boundary conditions. Journal of the mechanical behavior of biomedical materials Haese, C. E., Dubey, V., Mathur, M., Pouch, A. M., Timek, T. A., Rausch, M. K. 2024; 163: 106879

  Abstract

  Transcatheter edge-to-edge repair (TEER) simulations may provide insight into this novel therapeutic technology and help optimize its use. However, because of the relatively short history and technical complexity of TEER simulations, important questions remain unanswered. For example, there is no consensus on how to handle the annular boundary conditions in these simulations. In this short communication, we tested the sensitivity of such simulations to the choice of annular boundary conditions using a high-fidelity finite element model of a human tricuspid valve. Therein, we embedded the annulus among elastic springs to simulate the compliance of the perivalvular myocardium. Next, we varied the spring stiffness parametrically and explored the impact on two key measures of valve function: coaptation area and leaflet stress. Additionally, we compared our results to simulations with a pinned annulus. We found that a compliant annular boundary condition led to a TEER-induced "annuloplasty effect," i.e., annular remodeling, as observed clinically. Moreover, softer springs led to a larger coaptation area and smaller leaflet stresses. On the other hand, pinned annular boundary conditions led to unrealistically high stresses and no "annuloplasty effect." Furthermore, we found that the impact of the boundary conditions depended on the clip position. Our findings in this case study emphasize the importance of the annular boundary condition in tricuspid TEER simulations. Thus, we recommend that care be taken when choosing annular boundary conditions and that results from simulations using pinned boundaries should be interpreted with caution.

  View details for DOI 10.1016/j.jmbbm.2024.106879

  View details for PubMedID 39742687

• Leaflet remodeling reduces tricuspid valve function in a computational model JOURNAL OF THE MECHANICAL BEHAVIOR OF BIOMEDICAL MATERIALS Mathur, M., Malinowski, M., Jazwiec, T., Timek, T. A., Rausch, M. K. 2024; 152: 106453

  Abstract

  Tricuspid valve leaflets have historically been considered "passive flaps". However, we have recently shown that tricuspid leaflets actively remodel in sheep with functional tricuspid regurgitation. We hypothesize that these remodeling-induced changes reduce leaflet coaptation and, therefore, contribute to valvular dysfunction. To test this, we simulated the impact of remodeling-induced changes on valve mechanics in a reverse-engineered computer model of the human tricuspid valve. To this end, we combined right-heart pressures and tricuspid annular dynamics recorded in an ex vivo beating heart, with subject-matched in vitro measurements of valve geometry and material properties, to build a subject-specific finite element model. Next, we modified the annular geometry and boundary conditions to mimic changes seen in patients with pulmonary hypertension. In this model, we then increased leaflet thickness and stiffness and reduced the stretch at which leaflets stiffen, which we call "transition-λ." Subsequently, we quantified mean leaflet stresses, leaflet systolic angles, and coaptation area as measures of valve function. We found that leaflet stresses, leaflet systolic angle, and coaptation area are sensitive to independent changes in stiffness, thickness, and transition-λ. When combining thickening, stiffening, and changes in transition-λ, we found that anterior and posterior leaflet stresses decreased by 26% and 28%, respectively. Furthermore, systolic angles increased by 43%, and coaptation area decreased by 66%; thereby impeding valve function. While only a computational study, we provide the first evidence that remodeling-induced leaflet thickening and stiffening may contribute to valvular dysfunction. Targeted suppression of such changes in diseased valves could restore normal valve mechanics and promote leaflet coaptation.

  View details for DOI 10.1016/j.jmbbm.2024.106453

  View details for Web of Science ID 001180594300001

  View details for PubMedID 38335648

  View details for PubMedCentralID PMC11048730

• Geometric data of commercially available tricuspid valve annuloplasty devices DATA IN BRIEF Haese, C. E., Mathur, M., Malinowski, M., Timek, T. A., Rausch, M. K. 2024; 52: 110051

  Abstract

  Tricuspid valve annuloplasty is the gold standard surgical treatment for functional tricuspid valve regurgitation. During this procedure, ring-like devices are implanted to reshape the diseased tricuspid valve annulus and to restore function. For the procedure, surgeons can choose from multiple available device options varying in shape and size. In this article, we provide the three-dimensional (3D) scanned geometry (*.stl) and reduced midline (*.vtk) of five different annuloplasty devices of all commercially available sizes. Three-dimensional images were captured using a 3D scanner. After extracting the surface geometry from these images, the images were converted to 3D point clouds and skeletonized to generate a 3D midline of each device. In total, we provide 30 data sets comprising the Edwards Classic, Edwards MC3, Edwards Physio, Medtronic TriAd, and Medtronic Contour 3D of sizes 26-36. This dataset can be used in computational models of tricuspid valve annuloplasty repair to inform accurate repair geometry and boundary conditions. Additionally, others can use these data to compare and inspire new device shapes and sizes.

  View details for DOI 10.1016/j.dib.2024.110051

  View details for Web of Science ID 001171308700001

  View details for PubMedID 38299102

  View details for PubMedCentralID PMC10828561