Actin Filament Tracking Based on Particle Filters and Stretching Open Active Contour Models

Hongsheng Li, Tian Shen, Dimitrios Vavylonis, Xiaolei HuangMICCAI 2009 · OpenReview

Abstract

We introduce a novel algorithm for actin filament tracking and elongation measurement. Particle Filters (PF) and Stretching Open Active Contours (SOAC) work cooperatively to simplify the modeling of PF in a one-dimensional state space while naturally integrating filament body constraints to tip estimation. Our algorithm reduces the PF state spaces to one-dimensional spaces by tracking filament bodies using SOAC and probabilistically estimating tip locations along the curve length of SOACs. Experimental evaluation on TIRFM image sequences with very low SNRs demonstrates the accuracy and robustness of this approach.

Introduction

Actin filaments are essential proteins that drive critical cellular processes such as movement and division. To understand their kinetics—specifically how fast they grow and fluctuate—biologists utilize Total Internal Reflection Fluorescence Microscopy (TIRFM). However, TIRFM images are notoriously difficult to analyze automatically due to very low Signal-to-Noise Ratios (SNR). In these environments, the "tips" of the filaments, where growth occurs, often blend into random background noise. Previous automated methods frequently struggled with "over-growing" or "under-growing" errors and a lack of temporal coherency, often failing to maintain accurate tracks across consecutive frames.

Methodology

The researchers developed a cooperative framework that combines two distinct computer vision techniques to overcome these noise challenges.

Stretching Open Active Contours (SOAC)

Instead of attempting to find the exact tip immediately, the algorithm uses SOACs to model the filament's body. The SOAC is intentionally allowed to "over-grow" the filament, creating a reliable mathematical representation of the filament's path even if it extends slightly past the actual ends. This provides a stable "road" upon which the search for the tip can be conducted.

1D Particle Filtering

The core innovation of this work is the reduction of the search space. While standard Particle Filters track objects in 2D or higher-dimensional spaces—which is computationally expensive and error-prone in high noise—this method reduces the state space to a single dimension: the length along the SOAC curve. By searching only along this predefined path, the tracker becomes significantly more robust against background distractions.

SOAC models applied to single filaments (a, c) and crossing filaments (b) to define the 1D search space.

Biological Priors and Occlusion Handling

The model incorporates biological knowledge, such as the differing growth rates of the "Barbed" (B-end) and "Pointed" (P-end) tips. To handle filaments crossing over each other, a "confidence map" is used to lower the weight of pixels at intersections, preventing the tracker from jumping between filaments. Additionally, the Iterative Closest Point (ICP) algorithm is used for frame-to-frame registration to correct for global microscope stage drift.

Results and Discussion

The algorithm was evaluated using two TIRFM image sequences of muscle Mg-ADP-actin polymerization. Sequence I consisted of 20 images at 30-second intervals, while Sequence II included 34 images at 10-second intervals. Expert manual labeling of filament tips provided the ground truth for benchmarking.

Final tracking results on a low-SNR TIRFM image sequence, demonstrating robust multi-filament identification.

The proposed method showed a clear performance leap over previous intensity-based methods. In Sequence I, the mean tracking error dropped from 1.67 pixels to 0.88 pixels, with zero tracking failures. Even in the more complex Sequence II, the method maintained a lower mean error (1.73 vs. 2.90 pixels) and fewer total failures. The results confirm that constraining the probabilistic search to a 1D path along the filament body makes the tracker much more "sticky" and less likely to be distracted by noise.

Conclusion

By integrating the structural constraints of SOACs with the probabilistic power of Particle Filters, this algorithm provides a robust solution for tracking actin growth in challenging imaging conditions. While the system currently requires initial estimates of elongation rates and some manual oversight for initial frame segmentation, it represents a significant advancement in automated biological image analysis. Future work aims to extend this approach to dense actin networks where individual filaments are more difficult to distinguish.

Reference: Li, H., Shen, T., Vavylonis, D., & Huang, X. (2009). Actin Filament Tracking Based on Particle Filters and Stretching Open Active Contour Models. MICCAI 2009, Part II, LNCS 5762, pp. 673–681.

Summary updated

Abstract

We introduce a novel algorithm for actin filament tracking and elongation measurement. Particle Filters (PF) and Stretching Open Active Contours (SOAC) work cooperatively to simplify the modeling of PF in a one-dimensional state space while naturally integrating filament body constraints to tip estimation. Our algorithm reduces the PF state spaces to one-dimensional spaces by tracking filament bodies using SOAC and probabilistically estimating tip locations along the curve length of SOACs. Experimental evaluation on TIRFM image sequences with very low SNRs demonstrates the accuracy and robustness of this approach.
BibTeX
@inproceedings{li2009actin,
  author = {Hongsheng Li and Tian Shen and Dimitrios Vavylonis and Xiaolei Huang},
  title = {{Actin Filament Tracking Based on Particle Filters and Stretching Open Active Contour Models}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2009},
  url = {https://openreview.net/forum?id=d1ihxOAPiPk},
}
Figure