Artificial Intelligence in Digital Holographic Imaging

22Automated Analysis of Cardiomyocytes with Deep Learning

22.1 Introduction

Many efforts have been made to develop complementary methods for human‐induced pluripotent stem cell–derived cardiomyocytes (HiPSC‐CMs) characterization to reduce drug development costs and cardiotoxicity‐related drug attrition. CM characterization methods including patch clamping [1, 2], calcium imaging [3, 4], and image processing–based contraction‐relaxation [5, 6] are reported. The mechanical probe is another widely used method for CM characterization [7]. There are drawbacks associated with each method that require either expertise and costly equipment or plating CMs onto specialized material, which makes the process difficult. Furthermore, these methods were only applied to the entire slide images of multiple CMs, so CM characterization methods at the single‐cell level are lacking. As a result, it is important to exploit high‐throughput and reliable methods for single‐cell CM characterization.

In Chapter 21, we employed DHM for non‐invasive, label‐free studies of HiPSC‐CMs and subsequent beating activity quantification [8]. We showed that the nucleus section of HiPSC‐CMs from time‐lapse DHM clearly reflects its rhythmic beating pattern, which might be less noisy and more informative for subsequent characterization. The CM beating activity at the single‐cell level can thus be efficiently characterized if the dry‐mass redistribution signal is observed only in the CM nucleus region.

To analyze the ...