Glofinder: Ai-Empowered Qupath Plugin for Wsi-Level Glomerular Detection, Visualization, and Curation

Abstract

Recent advances in medical imaging have underscored the superiority of circle-based object detection techniques for identifying spherical structures, such as glomeruli, cells, and nuclei, compared to traditional bounding box-based methods. Circle representations naturally conform to the inherent geometry of these spherical entities, resulting in more accurate and meaningful localization. In typical bounding box-based detection tasks, combining predictions from multiple models has become a prevalent strategy to enhance accuracy, particularly in contexts where real- time processing constraints are relaxed. However, such ensemble approaches have yet to be effectively translated to circle-based detections, presenting a noticeable gap in existing methodologies. To address this limitation, our study introduces Weighted Circle Fusion (WCF), a novel ensemble method specifically tailored for merging circle-based predictions. WCF utilizes the confidence scores assigned to each predicted circle, enabling a weighted combination of outputs from multiple detection models, thereby significantly enhancing detection accuracy and reducing false positives. We validated the effectiveness of WCF using a proprietary dataset dedicated to glomerular detection within whole slide images (WSIs). Our results clearly demonstrate that the proposed method delivers superior performance in terms of precision and robustness compared to individual model predictions and conventional ensemble methods. Additionally, our investigation extended to evaluating annotation strategies, particularly exploring the benefits of incorporating a human-in-the-loop (HITL) approach. By integrating automated model predictions with human verification, we confirmed that HITL markedly accelerates annotation speed and improves data quality over traditional fully manual annotation methods. Capitalizing on these findings, we implemented WCF into a practical, user-friendly QuPath plugin named GloFinder, streamlining accurate glomeruli detection in WSIs and significantly optimizing medical imaging workflows.