Weighted Circle Fusion: Ensembling Circle Representation from Different Object Detection Results

Area of Science:

• Medical Imaging
• Computer Vision
• Pathology

Background:

• Circle representation aids spherical object identification in medical imaging.
• Ensemble methods improve bounding box detection but are not readily available for circle representations.
• Accurate detection of glomeruli is crucial for pathological analysis.

Purpose of the Study:

• To introduce Weighted Circle Fusion (WCF), a novel method for merging circle detection model predictions.
• To evaluate WCF's performance in glomerular detection within whole slide imaging (WSI).
• To compare the efficiency of manual versus human-in-the-loop (HITL) annotation for glomeruli.

Main Methods:

• Developed Weighted Circle Fusion (WCF) to merge circle predictions using confidence scores.
• Applied WCF to a proprietary dataset for glomerular detection in WSI.
• Assessed annotation efficiency using fully manual and HITL approaches for 200,000 glomeruli.

Main Results:

• WCF achieved a 5% performance gain over existing ensemble methods for glomerular detection.
• The HITL annotation approach significantly improved labeling efficiency compared to manual methods.
• WCF enhanced object detection precision and reduced false detections.

Conclusions:

• Weighted Circle Fusion offers a simple yet effective approach for combining circle detection models.
• HITL annotation significantly boosts efficiency in large-scale medical image labeling.
• WCF shows promise for improving pathological image analysis and reducing diagnostic errors.