CenFind: a deep-learning pipeline for efficient centriole detection in microscopy datasets.

Area of Science:

• Cell Biology
• Microscopy and Imaging
• Computational Biology

Background:

• Accurate centriole detection is crucial for understanding cellular processes but challenging in immunofluorescence images.
• Manual cell scoring is time-consuming, irreproducible, and lacks throughput.
• Existing automated methods often detect centrosomes, not centrioles, or require multi-channel inputs.

Purpose of the Study:

• To develop an efficient and versatile pipeline for automatic centriole detection in single-channel immunofluorescence datasets.
• To enable accurate and reproducible quantification of centriole numbers per cell.
• To address the unmet need for channel-intrinsic centriole detection.

Main Methods:

• Developed CenFind, a deep-learning pipeline utilizing the SpotNet multi-scale convolution neural network.
• Trained and evaluated CenFind on a custom dataset of human cell immunofluorescence images.
• Integrated a StarDist-based nucleus detector to link detected centrioles to individual cells.

Main Results:

• CenFind achieved an average F1-score of over 90% on the test set, demonstrating high accuracy and robustness.
• The pipeline enables automatic scoring of centriole numbers per cell from single-channel images.
• Demonstrated accurate detection of sparse and minute centriole foci in high-resolution images.

Conclusions:

• CenFind provides an efficient, accurate, and reproducible solution for centriole detection, filling a critical methodological gap.
• The pipeline is channel-intrinsic, overcoming limitations of existing multi-channel dependent methods.
• CenFind's modular design facilitates integration into other research pipelines, accelerating discoveries in cell biology.