IO-YOLO: A Feature-Enhanced YOLOv11 Model for Accurate Intestinal Organoid Staging

Intestinal organoids are fundamental three-dimensional platforms for investigating gut biology and pathology, but they present a significant analytical challenge. The conventional method of morphological assessment is time-consuming and susceptible to observer bias, which impedes progress in high-throughput investigations that depend on precise developmental staging. In this study, we developed and validated IntestiOrgano-YOLO (IO-YOLO), a computer vision framework designed to consistently classify murine intestinal organoid morphologies. Our methodology enhances the YOLOv11 framework by incorporating a customised Feature Enhancement Module (FEM) into its architecture. Based on multi-scale feature representation principles, this module uses a parallel-branch convolutional architecture with varied receptive fields and dilation rates. This design is theoretically grounded in augmenting the model's ability to discern the fine-grained visual cues that define distinct developmental stages. We conducted a systematic benchmark of IO-YOLO against the established YOLOv5n and YOLOv8n models, as well as the YOLOv11 baseline, using the public Tellu dataset under a stringent data-splitting protocol. On an independent test set, our model achieved an [email protected] of 0.832 and an [email protected]:0.95 of 0.553. These results signify a considerable performance improvement, with gains of 2.59% and 0.55% in [email protected] and [email protected]:0.95 respectively over standard YOLOv11. Further examination of F1-confidence curves and confusion matrices confirmed the superior diagnostic accuracy of our approach. We observed significant improvements in recognising spheroid and late-budding morphologies, which demonstrates the FEM's efficacy. We therefore propose IO-YOLO as a potent, validated instrument for advancing automated, high-precision morphological screening in intestinal organoid research.