Circle Representation for Medical Object Detection.

Area of Science:

• Computer Vision
• Medical Image Analysis
• Biomedical Engineering

Background:

• Object detection commonly uses bounding boxes, which are not optimal for spherical biomedical structures.
• Accurate detection of structures like glomeruli is crucial in renal pathology.
• Existing methods may lack robustness to rotation in medical imaging.

Purpose of the Study:

• To propose a novel circle representation for medical object detection.
• To introduce CircleNet, an anchor-free framework utilizing this circle representation.
• To evaluate the performance of circle representation against bounding boxes for biomedical objects.

Main Methods:

• Developed a bounding circle representation optimized for ball-shaped biomedical objects.
• Introduced CircleNet, an anchor-free object detection framework.
• Compared circle and bounding box representations on pathological images for glomeruli and nuclei detection.

Main Results:

• The proposed circle representation demonstrated superior detection performance for glomeruli and nuclei.
• Circle representation showed enhanced rotation invariance compared to bounding boxes.
• CircleNet achieved better accuracy and robustness in medical object detection tasks.

Conclusions:

• Circle representation is more suitable for detecting spherical biomedical objects than bounding boxes.
• CircleNet provides an effective and rotation-invariant solution for medical object detection.
• This approach advances the field of computer-aided diagnosis in pathology.