Abstract LB107: Use of high-speed multi-camera array microscopy and development of a novel machine learning model for high-throughput classification of patient-derived gastrointestinal organoids Free

Patient-derived organoids (PDOs) are a valuable tool for investigations of intra-tumor and inter-site heterogeneity and patient-specific drug responsiveness. PDOs as a preclinical and clinical model have immense potential in driving personalized medicine and targeted treatments but their use is hampered by many common impediments such as growth condition optimization and understanding organoid morphology efficiently early in organoid development. To address this we employ the MCAM (Multi-Camera Array Microscope) Vireo™ system to rapidly acquire brightfield images of PDOs derived from gastrointestinal surgical resection samples in under 2 minutes per 24 well plate and a new machine learning model to automatically analyze this data.

We used the MCAM Vireo to rapidly scan 24 well plates with patient derived PDO lines in brightfield at a 1.1um per pixel resolution. Images were preferentially acquired for paired tumor and “normal” tissue samples from the same patient, with data collected over several weeks, and acquisitions occurring before organoid passaging. Acquired images were used for development of a machine learning model for quantification of relevant organoid features from brightfield images. Brightfield data for PDO lines from initial tissues (pancreas and colorectal surgical resections) were split into training and testing datasets for machine learning model development. The training dataset images were labeled using a custom built micro-Segment Anything Model (SAM) based labeling tool and assigned relevant category labels. The model successfully identifies and segments organoids across tissue type and malignancy state, providing measures of growth and relevant morphology features such as circularity and eccentricity. Designations by the model are compared against histological data for each PDO line and pathologist classifications to ensure model validity.

The flexibility of this model to expand across new PDO lines and tissue types, paired with fast imaging and analysis using the MCAM Vireo, will enable researchers to evaluate the growth and phenotypes of PDOs at a speed and throughput that has not been possible until now. These tools overcome many of the acquisition and analysis throughput challenges and early classification limitations that have plagued the field, opening the door to larger scale use of PDOs. The early information provided by this model can inform decisions on which PDO cultures to propagate for additional testing and which PDO cultures to terminate, with image data and analysis to back up decision making rather than subjective and potentially variable selections.