A semi-automated algorithm for image analysis of respiratory organoids

TL;DR

A new AI tool helps scientists quickly and accurately analyze images of 3D respiratory organoids, improving research on diseases like cystic fibrosis.

Contribution

A semi-automated image analysis algorithm using U-Net and CellProfiler for respiratory organoid segmentation, validated with a public dataset.

Findings

The algorithm achieved high accuracy (IoU 0.8856, F1-score 0.937) in segmenting respiratory organoid images.

It successfully quantified CFTR-channel activity differences in cystic fibrosis organoids without fluorescent dyes.

An open-source dataset of 827 annotated organoid images was provided to support future research.

Abstract

Respiratory organoids have emerged as a powerful in vitro model for studying respiratory diseases and drug discovery. However, the high-throughput analysis of organoid images remains a challenge due to the lack of automated and accurate segmentation tools. This study presents a semi-automatic algorithm for image analysis of respiratory organoids (nasal and lung organoids), employing the U-Net architecture and CellProfiler for organoids segmentation. The algorithm processes bright-field images acquired through z-stack fusion and stitching. The model demonstrated a high level of accuracy, as evidenced by an intersection-over-union metric (IoU) of 0.8856, F1-score = 0.937 and an accuracy of 0.9953. Applied to forskolin-induced swelling assays of lung organoids, the algorithm successfully quantified functional differences in Cystic Fibrosis Transmembrane conductance Regulator (CFTR)-channel activity between healthy donor and cystic fibrosis patient-derived organoids, without fluorescent dyes. Additionally, an open-source dataset of 827 annotated respiratory organoid images was provided to facilitate further research. Our results demonstrate the potential of deep learning to enhance the efficiency and accuracy of high-throughput respiratory organoid analysis for future therapeutic screening applications.

In this study, we developed a semi-automated tool to analyze images of respiratory organoids—3D cell structures that mimic the human respiratory system. These organoids are vital for studying diseases like cystic fibrosis and testing potential drugs, but manually analyzing their images is time-consuming and prone to errors. Our tool uses artificial intelligence (AI) to quickly and accurately measure organoid size and shape from bright-field microscope images, eliminating the need for fluorescent dyes that can harm cells. We trained our AI model on a publicly shared dataset of 827 annotated organoid images, achieving high accuracy in detecting and quantifying organoids. When applied to cystic fibrosis research, the tool successfully measured differences in organoid swelling (forskolin-induced swelling - a key test for drug response) between healthy and patient-derived samples. By making our dataset and method openly available, we hope to support further research into respiratory diseases. Our work bridges the gap between complex lab techniques and practical applications, offering a faster, more reliable way to study human health and disease.

Figures