An AI-Based Automated Measurement Tool for the Morphological Characterization of Intestinal Organoids

The automated morphological analysis of intestinal organoids is a critical measurement for evaluating epithelial regeneration, disease modeling, and drug efficacy in personalized medicine. As these 3D models provide vital insights into tissue dynamics, traditional manual inspection has become a bottleneck; it is time-consuming, subjective, and prone to significant observer bias, which severely limits experimental scalability. This paper presents an automated measurement tool based on Artificial Intelligence for the multi-class region detection and morphological state labeling of organoids in microscopy images. We compare the performance of YOLOv10 as a standalone measurement system against a hybrid pipeline that utilizes YOLOv10 for region-of-interest (ROI) identification and ResNet50 for high-dimensional feature extraction. To identify the optimal decision-making layer, we evaluated and compared several Machine Learning classifiers, including Logistic Regression, Random Forest, and XGBoost. The models were validated on an extensive dataset of 840 images containing over 23,000 manually annotated objects, representing a wide spectrum of morphological phenotypes. The hybrid pipeline demonstrated superior measurement precision over the standalone approach. Specifically, the integration with Logistic Regression outperformed other classifiers, achieving Area Under the Curve (AUC) values up to 0.98. This high performance indicates that the hybrid architecture effectively captures subtle structural nuances and complex features that are often missed by single-stage detectors. The proposed tool ensures exceptional reproducibility and accuracy for high-throughput monitoring of 3D biosystems. By eliminating human error and providing a scalable solution, this AI-driven approach facilitates the standardization of organoid-based assays, accelerating progress in both academic research and industrial drug discovery workflows.