Mathematical Imaging Methods for Mitosis Analysis in Live-Cell Phase Contrast Microscopy

Joana Sarah Grah, Jennifer Harrington, Siang‐Boon Koh, Jeremy A. Pike, A. Schreiner, Martin Burger, Carola‐Bibiane Schönlieb, Stefanie Reichelt

Introduction

In cancer research, monitoring cell division (mitosis) is essential for evaluating drug efficacy. While fluorescence microscopy provides high contrast, the required markers and laser exposure are often phototoxic, inducing cellular stress that can alter experimental outcomes. Phase contrast microscopy offers a non-invasive alternative, but it introduces significant optical artifacts—specifically the "halo" and "shade-off" effects—that complicate automated analysis. This paper introduces a robust mathematical framework and a MATLAB-based tool, MitosisAnalyser, to automate mitosis detection and tracking without fluorescent labels.

Overcoming Phase Contrast Artifacts

Phase contrast images are notoriously difficult for computer vision. The "halo effect" creates bright rings around membranes, while "shade-off" causes cell interiors to blend into the background intensity. To bypass these issues, the authors utilized the local standard deviation of pixel values rather than simple intensity gradients. This approach captures the distinct texture of the cell interior, providing a more reliable signal for edge detection in noisy phase contrast environments.

Detection and Tracking Workflow

The pipeline begins by identifying cells as they enter mitosis and "round up." This morphological change is detected using the Circular Hough Transform (CHT), an algorithm optimized for finding circular shapes in noisy images by transforming edge points into a parameter space where they "vote" for potential circle centers.

Conceptual diagram of the Circular Hough Transform (CHT) used to detect rounded mitotic cells.

Following detection, the framework employs variational level-set methods. This involves a mathematical surface that evolves over time to precisely "shrink-wrap" the cell boundary. A unique "backwards-and-forwards" tracking strategy is then applied: the algorithm tracks the cell into the past to find the exact onset of mitosis and into the future to determine the cell's fate—whether it successfully divides or undergoes apoptosis (cell death).

The MitosisAnalyser Tool

The complex mathematical operations are encapsulated in MitosisAnalyser, a user-friendly MATLAB GUI designed for biological researchers. The tool allows for the extraction of high-quality kinetic data, such as Average Mitosis Duration (AMD) and cell fate classification, through an accessible interface that visualizes tracking results and statistical trends.

The MitosisAnalyser MATLAB interface, showing automated cell tracking and kinetic statistics.

Results and Validation

The framework was validated using MIA PaCa-2, HeLa Aur A, and T24 cancer cell lines. Key results include:

  • High Accuracy: A 93.3% accuracy rate in classifying cell fates (normal division vs. apoptosis).
  • Precise Timing: Measured mitosis durations (approx. 51 minutes) aligned perfectly with manual human scoring.
  • Robustness: The system successfully tracked complex "multipolar" divisions, where a single cancer cell splits into three or more daughters.

Conclusion

By combining morphology-based detection with advanced variational tracking, this workflow provides a powerful, label-free method for studying cell cycle kinetics. Future work aims to automate parameter tuning through machine learning and enable real-time analysis during live-cell experiments.

Summary updated

BibTeX
@inproceedings{grah2017mathematical,
  author = {Joana Sarah Grah and Jennifer Harrington and Siang‐Boon Koh and Jeremy A. Pike and A. Schreiner and Martin Burger and Carola‐Bibiane Schönlieb and Stefanie Reichelt},
  title = {{Mathematical Imaging Methods for Mitosis Analysis in Live-Cell Phase Contrast Microscopy}},
  booktitle = {Methods},
  year = {2017},
  doi = {10.1016/j.ymeth.2017.02.001},
}
Figure