SpindlesTracker

SpindlesTracker quantifies mitotic spindle dynamics from fluorescence microscopy by detecting and tracking spindle endpoints to measure spindle elongation.

Key Features:

• Automatic Detection and Tracking: Uses the YOLOX-SP neural network to detect spindle locations and endpoints with box-level data supervision.
• Optimized Algorithms: Enhances SORT (Simple Online and Realtime Tracking) and MCP (Multi-Channel Processing) for improved tracking and skeletonization of spindles.
• Low-Cost Data Labeling: Employs box-level supervision to reduce data labeling costs by 60%.
• Performance Metrics: Reports a mean average precision (mAP) of 84.1% for spindle detection, over 90% accuracy for endpoint detection, increases in tracking accuracy by 1.3% and tracking precision by 6.5%, and a mean spindle length error within 1 μm.
• Robustness to Noisy Image Sequences: Designed to operate on noisy fluorescence microscopy image sequences and complex backgrounds.
• Extensibility: Methodology can be extended to analyze other filamentous structures.

Scientific Applications:

• Mitotic spindle dynamics analysis: Quantitative analysis of spindle elongation and endpoint behavior during mitosis from fluorescence microscopy data.
• S.pombe spindle studies: Analysis and evaluation using a novel S.pombe dataset annotated from real-world acquisitions.
• Noisy experimental datasets: Application to fluorescence microscopy experiments with complex or noisy backgrounds.
• Filamentous structure analysis: Extension of the approach to study other filamentous biological structures.

Methodology:

Employs the YOLOX-SP neural network with box-level data supervision and optimized SORT and MCP algorithms, developed and evaluated on a novel S.pombe dataset annotated from real-world fluorescence microscopy acquisitions.