Deep learning for label-free nuclei detection from implicit phase information of mesenchymal stem cells.

Zhengyun Zhang,Andrea Ravasio,Kim Whye Leong,Krystyn Van Vliet,George Barbastathis

doi:10.1364/boe.420266

Abstract

Monitoring of adherent cells in culture is routinely performed in biological and clinical laboratories, and it is crucial for large-scale manufacturing of cells needed in cell-based clinical trials and therapies. However, the lack of reliable and easily implementable label-free techniques makes this task laborious and prone to human subjectivity. We present a deep-learning-based processing pipeline that locates and characterizes mesenchymal stem cell nuclei from a few bright-field images captured at various levels of defocus under collimated illumination. Our approach builds upon phase-from-defocus methods in the optics literature and is easily applicable without the need for special microscopy hardware, for example, phase contrast objectives, or explicit phase reconstruction methods that rely on potentially bias-inducing priors. Experiments show that this label-free method can produce accurate cell counts as well as nuclei shape statistics without the need for invasive staining or ultraviolet radiation. We also provide detailed information on how the deep-learning pipeline was designed, built and validated, making it straightforward to adapt our methodology to different types of cells. Finally, we discuss the limitations of our technique and potential future avenues for exploration.

Highlights

Mesenchymal stem cells (MSCs) [1] are undifferentiated stromal cells commonly derived from bone marrow and adipose tissue
We found the closest nuclei position in the second set for each nuclei position in the first set, and vice versa, retaining only mutual matches whose Cartesian distance was less than 30 pixels, which allowed for human error in the placement of nuclei centers as well as some movement of the cells between the live and fixed imaging phases in the MID2 and MID3 series
The pattern of total training times suggest that the preprocessing on the CPU for feeding the input data to the neural networks dominated the training time for simpler networks with fewer layers, whereas actual training on the GPU started playing a larger role with the bigger networks, as seen by the appreciable increase in training time starting with B8

Summary

Introduction

Mesenchymal stem cells (MSCs) [1] are undifferentiated stromal cells commonly derived from bone marrow and adipose tissue. Unlike pluripotent embryonic stem cells, MSCs are multipotent and are limited in their differentiation capacity—they mainly become osteocytes (bone), chondrocytes (cartilage) and adipocytes (fat). Despite this limitation, MSCs have been shown to hold great potential as a cell therapeutic agent [2], evident in the sheer number of ongoing clinical trials for a variety of diseases and conditions. In recent years, various approaches based on phase contrast [6] and holographic imaging [7,8] have been implemented to estimate cell numbers grown on flat culturing substrates and on microcarriers [9]

Objectives

Methods

Results

Conclusion