A semi-automated algorithm for image analysis of respiratory organoids

Area of Science:

• Biomedical Engineering
• Computational Biology
• Cell Biology

Background:

• Respiratory organoids are vital in vitro models for disease research and drug discovery.
• High-throughput analysis of organoid images is hindered by a lack of automated segmentation tools.

Purpose of the Study:

• To develop and validate a semi-automatic algorithm for accurate segmentation and analysis of respiratory organoid images.
• To enable efficient high-throughput screening for respiratory diseases and therapeutic development.

Main Methods:

• Utilized U-Net architecture and CellProfiler for semi-automatic segmentation of nasal and lung organoids.
• Processed bright-field images via z-stack fusion and stitching.
• Developed an open-source dataset of 827 annotated respiratory organoid images.

Main Results:

• Achieved high segmentation accuracy with an IoU of 0.8856, F1-score of 0.937, and overall accuracy of 0.9953.
• Successfully quantified functional Cystic Fibrosis Transmembrane conductance Regulator (CFTR)-channel activity in lung organoids without fluorescent dyes.
• Demonstrated the algorithm's utility in distinguishing CFTR activity between healthy and cystic fibrosis patient-derived organoids.

Conclusions:

• Deep learning significantly enhances the efficiency and accuracy of respiratory organoid image analysis.
• The developed algorithm and dataset support high-throughput screening for respiratory disease therapeutics.
• This approach offers a powerful tool for advancing respiratory medicine and drug discovery.