This paper presents a robust mathematical framework and a user-friendly MATLAB tool, MitosisAnalyser, designed to automate the detection and tracking of cell division (mitosis) in phase contrast microscopy. By combining the Circular Hough Transform for detection with variational level-set methods for tracking, the authors provide a way to extract high-quality kinetic data from live-cell imaging without the need for phototoxic fluorescent markers.
TL;DR
Researchers have developed a workflow that automatically detects and tracks the entire lifecycle of mitotic cells in phase contrast microscopy. By using "backwards-and-forwards" tracking and a custom edge-detection logic, the system accurately measures mitosis duration and cell fates (division vs. death) while avoiding the cellular stress caused by traditional fluorescent labeling.
What Problem Does This Paper Address?
In cancer research, scientists need to see how drugs affect cell division. While fluorescence microscopy makes cells easy to see, the lasers and chemical markers used can be "phototoxic," essentially stressing or killing the cells you are trying to study.
Phase contrast microscopy is a "gentle" alternative that requires no labels, but it is a nightmare for computer vision. It suffers from two major optical artifacts:
- The Halo Effect: Bright rings of light around cell membranes.
- The Shade-off Effect: The interior of a cell often has the same intensity as the background, making standard edge-detection algorithms fail.
Because of these issues, many researchers still analyze these images manually—a process that is slow, subjective, and prone to bias. This paper aims to bridge that gap with a fully automated mathematical workflow.
Methods
The authors propose a multi-step pipeline:
- Detection via Circular Hough Transform (CHT): When cells enter mitosis, they "round up" and become nearly spherical. The CHT is an algorithm specifically designed to find circular shapes in noisy images. It serves as the "trigger" to identify when and where a mitotic event begins.
- Variational Tracking (Level-Set Methods): Once a cell is detected, the algorithm uses a "level-set" function—a mathematical surface that evolves over time to "shrink-wrap" the cell's boundary.
- The "Backwards-Forwards" Strategy: This is a key innovation. To find the exact start of mitosis, the algorithm tracks the cell backwards in time from the point of detection until it loses its circular shape. Then, it tracks forwards to monitor the division into daughter cells or cell death (apoptosis).
- Custom Edge Detection: Instead of relying on simple light/dark gradients (which fail in phase contrast), the authors used the local standard deviation of pixel values. This captures the "texture" of the cell interior compared to the smooth background.
Datasets
The framework was validated using three different human cancer cell lines:
- MIA PaCa-2 (Pancreatic cancer): These cells expressed FUCCI (Fluorescent Ubiquitination-based Cell Cycle Indicator). The fluorescence was used only as a "ground truth" to verify that the phase-contrast-only algorithm was getting the timing right.
- HeLa Aur A (Cervical cancer): Modified to over-express certain kinases, used to test the algorithm's ability to handle abnormal divisions (e.g., one cell splitting into three).
- T24 (Bladder cancer): Used to test the classification of different cell fates under various drug treatments (Paclitaxel, Docetaxel, etc.).
Key Insights and Ideas
- Exploiting Morphology: The "aha moment" is realizing that the physical rounding of a cell is a more reliable computational signal than the messy intensity values of phase contrast images.
- Texture over Intensity: By using local variance (standard deviation) as an edge detector, the authors bypassed the "shade-off" effect where cell interiors look identical to the background.
- Temporal Context: Tracking in both directions (past and future) allows the software to reconstruct the entire mitotic history of a cell from a single detection point.
- Accessibility: The authors packaged the complex math into a MATLAB GUI called MitosisAnalyser, specifically designed for biologists who are not experts in imaging mathematics.
Key Results
- Accuracy: The algorithm achieved a 93.3% accuracy in classifying cell fates (normal division, apoptosis, or no division) in T24 cells.
- Timing: The measured Average Mitosis Duration (AMD) for control cells was approximately 51 minutes, which aligns perfectly with manual human scoring.
- Robustness: The tool successfully tracked "multipolar" divisions (cells splitting into 3+ daughters), a common occurrence in cancer cells treated with anti-mitotic drugs.
- Segmentation Quality: Using the Jaccard Similarity Coefficient (a measure of overlap), the proposed method significantly outperformed standard "off-the-shelf" algorithms like Chan-Vese or basic Geodesic Active Contours.
Limitations and Future Work
- Crowded Environments: While the tool handles touching cells reasonably well, extremely high cell densities can still confuse the tracking logic.
- Manual Parameter Tuning: Although the authors provide presets for different cell lines, some parameters (like expected cell radius) still require initial manual setup.
- Real-time Analysis: The current workflow is "off-line" (processed after the experiment). The authors suggest that future versions could run "on-line," allowing the microscope to automatically zoom in or change settings the moment a mitotic event is detected.
- Bilevel Optimization: The authors mention using machine learning to automatically "learn" the best mathematical parameters for new, unseen cell types.
Summary updated