ARGUS: Accelerated, Robust, General, and Unsupervised Cell Tracking Solutions

Automated cell tracking is essential for understanding biological dynamics, yet it remains computationally demanding due to image noise, cell overlap, and complex behaviors like division. The authors sought to develop a robust, efficient, and interpretable tracking framework that functions without the need for large training datasets or specialized GPU infrastructure.

ARGUS (Accelerated, Robust, General, and Unsupervised cell tracking Solutions) is a modular pipeline consisting of four stages:

1. Data Preparation: Standardizes images using intensity clipping and wavelet denoising to improve signal-to-noise ratios.
2. Cell Detection: Employs adaptive thresholding (Otsu’s method) to identify cell centroids.
3. Motion Estimation and Local Association: Uses Farneback optical flow to predict cell movement between frames, reducing the search space for linking detections via a linear assignment solver.
4. Global Tracklet Refinement: A final optimization step that reconnects trajectory fragments across short temporal gaps, ensuring temporal consistency and mitigating the impact of transient occlusions or missed detections.

The authors evaluated ARGUS against four datasets from the Cell Tracking Challenge (CTC). The framework achieved detection accuracy scores between 0.905 and 0.971 and tracking accuracy scores between 0.897 and 0.964. Notably, the system is highly efficient, processing frames in seconds without requiring GPU acceleration. Ablation studies confirmed that intensity clipping is the most critical preprocessing step for maintaining high accuracy, while the global refinement stage successfully reduced track fragmentation across all tested datasets.