Tracking Clathrin Coated Pits with a Multiple Hypothesis Based Method

Liang Liang, Hongying Shen, Pietro De Camilli, James S. DuncanMICCAI 2010 · DOI

Abstract

Automated tracking of Clathrin Coated Pits (CCPs) is essential for quantifying Clathrin-Mediated Endocytosis (CME). This study introduces a framework combining Poisson-noise-aware detection with a Multiple Hypothesis Tracking (MHT) algorithm. By incorporating biological priors such as constrained motion and intensity dynamics, the method significantly improves tracking fidelity in noisy, crowded microscopy environments, achieving performance comparable to human experts.

Introduction

In cell biology, Clathrin-Mediated Endocytosis (CME) is a fundamental process where cells internalize nutrients and signals. This process involves the formation of Clathrin Coated Pits (CCPs), which mature and eventually pinch off into the cell as vesicles. To study CME quantitatively, researchers must track thousands of these tiny, glowing spots across time-lapse image sequences.

The lifecycle of a Clathrin Coated Pit (CCP) from assembly to uncoating, alongside a typical fluorescent microscopy image of a cell.

The challenge in tracking CCPs is twofold. First, fluorescent microscopy is often plagued by "shot noise" (Poisson noise), making it difficult to distinguish real pits from background fluctuations. Second, CCPs are crowded; they appear, grow in brightness, move slightly, and then vanish. Standard trackers often assume particles move freely (Brownian motion) or have constant brightness, leading to broken or incorrectly merged trajectories.

Methodology

The proposed framework frames the problem as a Maximum A Posteriori (MAP) estimation task, divided into two main phases: detection and trajectory estimation.

Detection with Poisson Noise Modeling

Unlike many methods that assume Gaussian noise, this paper utilizes a Poisson noise model to better reflect the physics of photon counting in CCD cameras. The authors fit a Gaussian Mixture Model (GMM) to the image data to locate particles. To find optimal starting points for this optimization, they employ a combination of Laplacian of Gaussian (LoG) filtering and histogram-based thresholding to separate signal from noise.

Histogram-based thresholding (a, b) and the resulting detection of CCPs in microscopy images (c, d).

Biologically-Informed Trajectory Estimation

The core innovation lies in the motion and intensity models used during the data association (tracking) phase:

  • Constrained Brownian Motion: Recognizing that CCPs are tethered to the cell membrane, the authors propose a model where particles move randomly but stay within a restricted "leash" area.
  • Intensity Modeling: A linear Gaussian model describes how a CCP’s brightness typically increases during assembly and decreases as it prepares for fission.
  • Multiple Hypothesis Tracking (MHT): Instead of making a "greedy" decision at every frame, the MHT algorithm maintains several possible tracking scenarios over time, eventually picking the one that maximizes the overall probability.

Results and Validation

The method was validated using both synthetic and real-world datasets. On synthetic data containing 386 particles with varying Signal-to-Noise Ratios (SNR), the proposed method consistently outperformed standard trackers (e.g., Jaqaman et al.).

For real-world validation, TIRFM (Total Internal Reflection Fluorescence Microscopy) images of COS7 cells were used. A Kolmogorov-Smirnov test showed no significant difference ($p > 0.5$) between the automated trajectories and those produced by a human expert biologist. Notably, the method successfully reduced "zombie" links—erroneous trajectories created by linking newly appearing CCPs to recently vanished ones—by leveraging intensity and motion priors.

Conclusion

By integrating biological constraints directly into the tracking algorithm, this framework provides a robust solution for the quantitative analysis of CME. While the current implementation focuses on 2D membrane-bound processes, future work could extend these models to 3D tracking for studying vesicles moving deeper into the cytoplasm. The shift toward Poisson noise modeling and MHT represents a significant step forward in handling the inherent ambiguities of live-cell imaging.

Summary updated

BibTeX
@inproceedings{liang2010tracking,
  author = {Liang Liang and Hongying Shen and Pietro De Camilli and James S. Duncan},
  title = {{Tracking Clathrin Coated Pits with a Multiple Hypothesis Based Method}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2010},
  doi = {10.1007/978-3-642-15745-5\_39},
}
Figure