A review of bright-field and phase-contrast imaging finds different strengths and limits in rule-based, classical machine-learning and deep-learning analysis of spheroids and organoids. Dataset shifts, segmentation failures and two-dimensional projections remain practical concerns.
Three University of Pretoria researchers review automated analysis of bright-field and phase-contrast images from spheroids and organoids used in high-content screening. Rule-based approaches can standardize measurements that once required manual tracing, but may fail when imaging conditions change. Classical machine learning can handle some treatment-related changes in morphology, although it often needs retraining. Deep-learning tools learn from images more directly, yet their performance still depends on the training data and experimental setting.
The review recommends choosing tools according to the biological endpoint, image quality, culture shape, available annotations and computing resources. Most methods work from two-dimensional projections of three-dimensional structures, which can distort volume estimates for irregular spheroids or organoids. Performance also varies between data sets and imaging laboratories. The authors describe a choice of analytical approach tied to the assay and image set, rather than a universal ranking of software.
A Westlake University team trained an image-analysis framework with fluorescence-derived labels, then used bright-field images to estimate viability and death patterns. A 24-compound A2780 screen revealed responses that bulk ATP measurements alone did not distinguish.
Expression-based models trained on 2D cancer cultures retained a weak ranking signal in independent organoid cohorts. Their added value over tumor-lineage information in the blinded Sanger cohort remained uncertain.
