Imaging-Based Machine Learning Analysis of Patient-Derived Tumor Organoid Drug Response

Area of Science:

• Oncology
• Drug Discovery
• Bioimaging

Background:

• High failure rate of drug compounds in clinical trials necessitates improved preclinical screening.
• Patient-derived organoids offer a promising 3D model for drug discovery but present challenges in efficacy assessment.
• Traditional cell viability assays do not fully utilize the potential of organoid systems.

Purpose of the Study:

• To develop an image-based assay for evaluating drug efficacy in 3D patient-derived tumor organoids.
• To overcome limitations of traditional assays in phenotypically evaluating complex organoid models.
• To enable non-destructive, high-content analysis of individual organoid responses to drugs.

Main Methods:

• An image-based high-content assay utilizing computer vision for organoid segmentation and analysis.
• Acquisition of brightfield images at multiple timepoints for non-destructive, dynamic tracking.
• Development of a web-based tool (Organoizer) for simplified analysis of complex data.

Main Results:

• The assay provides object-level morphometric and textural information for individual organoids.
• Dynamic drug response of single organoids can be tracked non-destructively.
• Demonstrated feasibility of using imaging, computer vision, and machine learning to determine organoid vital status without vital dyes.

Conclusions:

• This novel assay enhances the utility of patient-derived organoids in preclinical drug discovery.
• The approach improves the translatability of compounds by providing detailed, individual organoid-level data.
• The developed methods and tools facilitate more effective drug screening and development in oncology.