Spotiflow: accurate and efficient spot detection for fluorescence microscopy with deep stereographic flow regression

Area of Science:

• Microscopy and Image Analysis
• Computational Biology
• Deep Learning Applications

Background:

• Accurate identification of spot-like structures is essential for life-science applications, particularly imaging-based spatial transcriptomics (iST).
• Current spot detection methods often rely on classical signal processing, requiring extensive manual tuning and struggling with noisy, low signal-to-noise images.
• Existing techniques lack efficiency and generalizability across diverse imaging conditions.

Purpose of the Study:

• To introduce Spotiflow, a novel deep learning method for highly accurate, subpixel-level spot detection in microscopy images.
• To address the limitations of traditional methods by providing a robust and automated solution for spot detection.
• To enhance the analysis of large-scale biological datasets, including iST and live imaging.

Main Methods:

• Spotiflow formulates spot detection as a multiscale heatmap and stereographic flow regression problem using deep learning.
• The method is designed to support both 2D and 3D image data.
• It is optimized for time and memory efficiency compared to existing approaches.

Main Results:

• Spotiflow achieves subpixel-accurate spot detection, outperforming traditional methods.
• The tool demonstrates generalizability across various imaging conditions and datasets.
• Quantitative experiments confirm the efficacy and accuracy of Spotiflow.

Conclusions:

• Spotiflow provides a significant advancement in automated spot detection for microscopy.
• Its improved accuracy enhances biological insights derived from iST and live imaging experiments.
• The method offers a time- and memory-efficient solution, available as a Python library and napari plugin.