Classifying Cardiac Fibroblasts with Predictive Modeling

Cardiac Fibroblasts (CF) are a type of connective cell in the heart that is known to play an integral role in cardiac fibrosis, a type of heart disease, following injury. Activated CFs are correlated with the onset and progression of disease. Recent cell studies have shown that various in vitro conditions lead to the activation of CFs. Traditionally, activated CFs are identified by visually inspecting the α-smooth muscle actin (aSMA) macrostructure through immunostaining. A more fibrous aSMA structure suggests the cells are in the activated state, while the lack of fibers suggests a non-activated state. In addition to displaying aSMA fibers, activated CFs have been correlated with larger cell sizes. A more quantitative analysis of these morphological trends is necessary to better understand the characteristics of CF cells and develop a model that classifies activated vs. non-activated CFs from a single cell population.

This poster presents a quantitative analysis of various morphological properties for CF cells in both the activated and non-activated states. Additionally, it demonstrates predictive modeling methods to automatically determine the activation state of CFs. Using fluorescent microscopy combined with computer vision techniques, we computed 15 unique morphological factors of the cells. These properties were fed into the various predictive modeling methods, with the best model reaching 91.2% accuracy.

Using the calculated morphological features, we demonstrate previously described unquantified trends in the literature. We have quantitatively determined that (a) activated and non-activated CF cells contain two unique populations with statistically different morphologies, and (b) activated cells contain a larger colocalization between the aSMA and actin structures compared to non-activated cells. With our computer vision predictive modeling methods, we significantly decrease the classification time of CF cells while minimizing scientist bias.