Object Stitching by Clustering of adjacent Regions designed for accurate quantification of three-dimensional tissues

The study of tissue organization and morphogenesis requires quantitative analysis of three-dimensional biological samples, a challenging task due to limitations in imaging dense organs at single-cell resolution. Current 3D segmentation and quantification tools often struggle with the low resolution and signal-to-noise ratios typical of images taken in vivo or deep within tissues. To address this, we developed OSCAR (Object Stitching by Clustering of Adjacent Regions), a framework that combines machine learning with nonlinear fitting and statistical algorithms specifically designed to quantify biological 3D stacks with high cellular density and low signal-to-background ratio based on nuclear staining. As proof of principle, the framework is applied to quantify the growth and organizational dynamics of the developing zebrafish vertebrate retina, showing that the tissue grows exponentially in cell numbers while cell density increases and average nuclear volume decreases. Overall, its high accuracy, ease of use, and reduced computational requirements establish OSCAR as a valuable tool for automated image analysis of densely packed tissues composed of cell subtypes that can be distinguished by specific labeling.