A deep learning approach to identify and segment alpha-smooth muscle actin stress fiber positive cells

Abstract

Cardiac fibrosis is a pathological process characterized by excessive tissue deposition, matrix remodeling, and tissue stiffening, which eventually leads to organ failure. On a cellular level, the development of fibrosis is associated with the activation of cardiac fibroblasts into myofibroblasts, a highly contractile and secretory phenotype. Myofibroblasts are commonly identified in vitro by the de novo assembly of alpha-smooth muscle actin stress fibers; however, there are few methods to automate stress fiber identification, which can lead to subjectivity and tedium in the process. To address this limitation, we present a computer vision model to classify and segment cells containing alpha-smooth muscle actin stress fibers into 2 classes (α-SMA SF+ and α-SMA SF-), with a high degree of accuracy (cell accuracy: 77%, F1 score 0.79). The model combines standard image processing methods with deep learning techniques to achieve semantic segmentation of the different cell phenotypes. We apply this model to cardiac fibroblasts cultured on hyaluronic acid-based hydrogels of various moduli to induce alpha-smooth muscle actin stress fiber formation. The model successfully predicts the same trends in stress fiber identification as obtained with a manual analysis. Taken together, this work demonstrates a process to automate stress fiber identification in in vitro fibrotic models, thereby increasing reproducibility in fibroblast phenotypic characterization.

A strategy to quantify myofibroblast activation on a continuous spectrum

“…While glass has been shown to activate cells similarly to chemical stimuli such as TGF-β1 48 , this activation is far from complete. Many cells remain non-activated and are visually indistinguishable from cells grown on a soft substrate, such as those seen in our previous work 29 . Thus, these samples represent a heterogeneous population.…”

“…More complex algorithms such as convolutional neural networks (CNNs) have also been used to identify and classify images at or above human performance 24 . In the biomedical field, these models have been developed for applications such as cell segmentation 25 , cell classification 26 , 27 , and tissue segmentation 28 , including our previous work to classify cardiac fibroblasts on the binary scale 29 . Lastly, recent advances in deep learning models have helped to remove human bias from scientific systems.…”

A Strategy to Quantify Myofibroblast Activation on a Continuous Spectrum

“…After cycling through all measured cell areas, the optimal cutoff value was determined to be approximately 8,000 μm 2 , which alone yields an accuracy of 76% when compared to our manual activation predictions based on the appearance of α-SMA stress fibers. This cell size is similar to that of activated cells cultured on stiff hydrogel substrates 28 . This result was promising, but still too inaccurate to be used to automate the classification of cell activation.…”

“…While glass has been shown to activate cells similarly to chemical stimuli such as TGF-β1 47 , this activation is far from complete. Many cells remain non-activated and are visually indistinguishable from cells grown on a soft substrate, such as those seen in our previous work 28 . Thus, these samples represent a heterogeneous population.…”

“…More complex algorithms such as convolutional neural networks (CNNs) have also been used to identify and classify images at or above human performance 23 . In the biomedical field, these models have been developed for applications such as cell segmentation 24 , cell classification 25,26 , and tissue segmentation 27 , including our previous work to classify cardiac fibroblasts on the binary scale 28 . Lastly, recent advances in deep learning models have helped to remove human bias from scientific systems.…”