StructOrg: Organoid Detection with Structural Feature Modeling and Edge Feature Extraction

Organoids are biomimetic three-dimensional \textit{in vitro} models vital for drug screening and biological research. However, conventional analysis paradigms reliant on manual annotation and fluorescent staining suffer from labor-intensive workflows, subjective bias, and cytotoxicity. To address these bottlenecks, we propose StructOrg, an ultra-lightweight framework designed for high-throughput, stain-free bright-field organoid detection. Specifically, StructOrg integrates a Grouped Shift-Conv Channeler (GSCC) into the downsampling stages, leveraging channel grouping, spatial shifts, and adaptive channel scaling to capture subtle structural deformations with minimal parameter overhead. In the detection head, an Edge Feature Extractor (EFE) combines differential image mechanisms with orthogonal skip-sampling to sharpen indistinct boundary representations. Comprehensive evaluations on a multi-class organoid dataset demonstrate that StructOrg consistently outperforms mainstream detectors in both detection accuracy and inference throughput, establishing a practical, deployable solution for automated bright-field organoid analysis.