CLORG: A contrastive learning-based framework for morphological representation and classification of organoids

Organoids are three-dimensional structures derived from stem cells or primary cells, widely used in disease research and regenerative medicine. However, the presence of significant noise and morphological heterogeneity in their bright-field images makes it challenging to distinguish between different categories of organoids. This study is the first to propose a deep learning framework, CLORG, based on supervised contrastive learning. By narrowing the distance between samples of the same class through contrastive learning and incorporating Fourier transform to enhance the representation of frequency-domain information, the framework efficiently performs multi-class classification of organoids. This, in turn, facilitates the analysis of organoid developmental trends and supports drug screening and evaluation. Experiments on colon and intestinal organoid datasets demonstrate that CLORG achieves accuracies of 91.68% and 86.93%, respectively, with improvements of 3.35% and 1.89% over baseline models. The findings validate the effectiveness of CLORG in organoid image multi-class classification tasks and highlight its significant implications for organoid analysis and research.