Machine Learning-Based Brightfield Detection of Anti-Fibrotic Effects in a 3D Retinal Fibrosis Model.

Three-dimensional (3D) spheroid and microtissue models are increasingly used in phenotypic drug discovery, yet their scalability for high-throughput screening often remains limited, constrained by the cost and complexity of conventional imaging-based readouts. Here, we present a morphology-based screening approach using low-resolution brightfield images of human retinal pigment epithelium microtissues (hRPE-MTs) to identify fibrotic and non-fibrotic phenotypes. The results of this study demonstrate that extracting geometric, intensity, and texture features from segmented spheroids enables quantitative characterization of fibrosis-associated morphological changes. Machine learning classifiers trained on these features effectively discriminated between fibrotic and non-fibrotic hRPE-MTs. Collectively, these findings indicate that label-free brightfield morphology encodes sufficient information to support scalable phenotypic screening in 3D fibrosis models. This approach provides a low-complexity and broadly accessible strategy for early-stage drug discovery.