Computational methods and open-source tools for spatial and temporal characterization of heterogeneous soft tissue
OA Version
Citation
Abstract
Over the past two decades, rapid advances in computational infrastructure and technology have created new opportunities for modeling the mechanics of biological soft tissues. To fully realize these opportunities, we need computational approaches that make soft tissue characterization more accurate, efficient, and scalable. Yet significant challenges remain. Broadly, these challenges fall into two categories: those that stem from the inherent complexity, heterogeneity, and intricate microstructure of biological tissues, and those that are more tractable, such as the lack of benchmark datasets, standardized metrics, open measurement tools, and robust data-sharing practices. This dissertation focuses on addressing these gaps using cardiac tissue engineering as the primary testbed, developing robust computational pipelines and data-driven methods for descriptive and diagnostic analyses. It also introduces openly available datasets and software tools that support reproducible and scalable characterization of engineered cardiac tissues. Together, these efforts are a step towards transforming the study of soft tissue mechanics from a labor-intensive and fragmented process into one that is automated, quantitative, and broadly accessible to the scientific community. In the first chapter of this dissertation, we draw inspiration from the transformative role of large-scale image datasets in computer vision to address the shortage of benchmark datasets for modeling the mechanics of heterogeneous soft tissues. To this end, we developed new resources and strategies for dataset generation and augmentation. We first created the Mechanical MNIST Cahn-Hilliard dataset, an open-access finite element simulation dataset comprising two-dimensional heterogeneous input patterns and their corresponding equibiaxial extension simulation results. Recognizing the limitations imposed by small datasets in tissue characterization, we then introduced a style-based generative adversarial network (GAN) with adaptive discriminator augmentation to generate realistic synthetic patterns from only 1,000 example inputs. These synthetic patterns were shown to be effective as inputs for finite element simulations, substantially expanding dataset diversity and utility. Alongside the dataset, we released the complete computational pipeline for data generation, augmentation, and analysis, providing an open and extensible resource for researchers advancing computational methods in material and tissue mechanics. In the remaining chapters of this dissertation, we focus on applications that directly advance the frontiers of soft tissue biomechanics through the study of human induced pluripotent stem cell (hiPSC)-derived cardiac microbundles, engineered heart tissues that serve as powerful platforms for disease modeling, drug discovery, and regenerative medicine. Despite their promise, a key challenge remains: non-destructively quantifying the dynamic beating behavior of these tissues in a consistent and high-throughput manner. To address this, we developed two open-source computational frameworks, "MicroBundleCompute" and "MicroBundlePillarTrack," which automate the extraction of morphology-based mechanical metrics from microscopy videos of lab-grown cardiac microtissues. "MicroBundleCompute" provides full-field displacement maps and subdomain-averaged strain measures within the tissue, while "MicroBundlePillarTrack" quantifies pillar deflection and the corresponding average tissue stress at attachment interfaces. Together, these tools transform what was once a manual, low-throughput process into a scalable, reproducible, and quantitative workflow for cardiac tissue analysis. Beyond enabling more rigorous cardiac biomechanics studies, these frameworks also open new interdisciplinary applications, such as characterizing global muscle sheet contraction under optical stimulation, thereby expanding the possibilities for computationally informed soft tissue research. Building on these computational tools, we developed advanced analytical methods to examine the spatiotemporal heterogeneity of engineered cardiac tissues. Specifically, we introduced new quantitative metrics that capture the structural organization, functional performance, and synchronization dynamics of lab-grown cardiac tissues. Using these metrics in combination with experimental data, we applied statistical and machine learning approaches to identify underlying patterns and provide a robust framework for analyzing complex biological datasets. This work not only reveals intricate biomechanical behaviors of engineered cardiac tissues but also establishes a rigorous and transparent foundation for data-driven investigation, helping to mitigate risks of overfitting and data dredging in quantitative cardiac research. Collectively, this dissertation contributes rigorously evaluated computational methods for characterizing biological soft tissues. The key scientific advances are: (1) new strategies for generating and augmenting limited heterogeneous material datasets with generative machine learning, (2) open-source frameworks for automated, high-throughput quantification of tissue mechanics, and (3) analytical approaches for probing spatiotemporal heterogeneity in engineered tissues. By addressing data scarcity, measurement limitations, and computational complexity, this work strengthens the methodological basis for systematic and reproducible studies of soft tissue biomechanics. Through the open release of datasets, software, and reproducible computational pipelines, this dissertation advances not only methodological innovation but also the principles of open, transparent, and collaborative science in soft tissue biomechanics.
Description
2026
License
Attribution 4.0 International