Thesis defense – Christer Lohk

Venue: IBGC

Christer Lohk
Team : Imagerie quantitative de la cellule (Sibarita)
IINS

Thesis supervised by Jean-Baptiste Sibarita (IINS) & Macha Nikolski ( IBGC)

Title

“Articificial intelligence for the characterization of 3D cultures in high content fluorescence microscopy”

Abstract

Over the recent years, 3D cell cultures have become the gold standard in cell biology, with applications ranging from fundamental research to high-content drug screening and biomedicine. Among them, organoids oVer a major advantage over traditional two-dimensional cell cultures as they accurately replicate the cellular architecture and function of a tissue of interest. 3D cell cultures have, therefore, become irreplaceable as biological models for studying disease mechanisms, normal tissue development, and screening drug responses.

The main focus of this thesis project was the development of computational tools for the analysis of 3D cell culture images obtained with the light sheet microscopy-based high- content screening platform developed in our lab. This acquisition system employs disposable microfabricated chips, containing arrays of small pyramidal wells enabling the cultivation and parallel three-dimensional imaging of individual organoids. With this approach, imaging data can be acquired simultaneously from over a hundred single organoids using two complementary modalities: i) fluorescence single-objective Selective Plane Illumination Microscopy (soSPIM), which captures three-dimensional multi-channel image stacks, and ii) transmission light modality, allowing for rapid, label-free acquisition of two-dimensional images of the samples. When used in conjunction with an integrated well detection algorithm, the image acquisition process can be fully automated and run over extended periods of time.

This thesis presents a toolkit of computational workflows for monitoring and quantification of drug-induced changes in spheroid and organoid cell cultures. Our proposed methods include i) a deep learning-based real-time segmentation model to rapidly localize organoid boundaries in transmission light images, ii) foundation model- based approach for volumetric segmentation of organoids in three-dimensional images, and iii) a feature extraction pipeline that integrates classical image features with deep representations from vision transformer and variational autoencoder networks to quantify morphological diVerences across conditions. The feature extraction strategy supports both quality control during acquisition workflow and comprehensive toolset for phenotypic profiling of organoids in fluorescence and brightfield images. Combined with the soSPIM-based culturing and imaging capabilities, these established methods provide a foundation for future large-scale, fully automated experimental pipelines targeted at predicting the response of 3D cell cultures to chemical and biological treatments.

Keywords

3D cell culture, image processing and analysis, fluorescence microscopy, deep learning

Jury

• WALTER Thomas, Directeur de recherche, Institut Curie, Examinateur
• ZIMMER Christophe, Professeur, Rudolf Virchow Center, University of Würzburg Rapporteur
• DESCOMBES Xavier, Directeur de recherche, INRIA Sophia Antipolis, CNRS, Rapporteur
• SIBARITA Jean-Baptiste, Chargé de recherche, Université de Bordeaux, CNRS, Directeur de Thèse
• NIKOLSKI Macha, Directrice de recherche, IBGC, CNRS, Université de Bordeaux, Co-Directrice de Thèse

• Dates
  On September 19, 2025 at 09:00