Methods for Analyzing Myofibril Orientation of hIPSC Derived Cardiomyocytes in 3D

Abstract

Human induced pluripotent stem cell derived cardiomyocytes (hiPSC-CMs) are used for studying the function of the heart, as well as cardiac disease and drug response in vitro. Even so, they have limitations as they do not mature to the level of a cardiomyocyte of adult human in vitro. This results in differences in the myofibril structure of hiPSC-CM and fully mature cardiomyocyte. While adult cardiomyocyte has linear and highly organized myofibril structure, that of hiPSC-CM is non-linear and irregular. The degree of organization varies significantly between hiPSC-CMs. Myofibrils are contractile structures, and therefore change in myofibril structure causes change in contractility and ultimately the beating of the cell. Due to this, knowing the myofibril structure is important for estimating cardiac function. While there are several methods for analyzing the myofibrils structure in 2D, there are no methods for doing the same in 3D to the author’s knowledge.

This bachelor’s thesis includes both literature review and experimental part. In the literature review, methods for analyzing fiber orientation in 3D were surveyed. As no existing method for hiPSC-CMs was found, the methods examined were for fibers in different applications. The results of the literature review reveal that there is no clear solution for hiPSC-CM myofibril orientation analysis in 3D. While using vector field to map the orientation describes the directionality of a curved fiber most accurately, it omits the information for a single fiber. This reduces the number of possible usages for acquired orientation data. Using azimuthal and polar angles to describe fiber orientation allows further calculations for the data. For example, long fibers can be emphasized by weighing each orientation with the length of the fiber. Fiber orientation for single fibers, however, is not ideal for non-linear fibers as one set of angles fails to consider the curve of the fiber.

The aim of the experimental part was to separate areas containing myofibrils from non-myofibril areas. For this purpose, YOLOv4-based convolutional neural network was used. The results of the experimental part show that the model could be used, for example, evaluating the usefulness of a larger dataset as a training material for a machine learning algorithm. Even so, it still needs improvement for reliably segmenting areas with myofibrils.