Fig. 9
From: Automatic ovarian follicle detection using object detection models
Errors analysis of expert and model identification of ovarian structures. This figure presents section of ovarian tissue with structures identified by both human expert (Blue bounding boxes) and AI model (Yellow bounding boxes). The expert categorized discrepancies into three types of errors: (1) The geniune or real errors (Red boxes) : structures identified by experts but missed by the model. These are clear structures that the model should have recognized but failed to do so, (2) Allowable erros (Light red boxes) : structures recognized by the expert but not labeled by the model due to staining inconsistencies, mounting issues, scanning artefacts, or other technical factors, and (3) Errors corrected by the model (Green Boxes) : structures overlooked by the expert due to factors such as eye fatigue or intentionally unlabelled (tears in the sections) but correctly identified by the AI model. These instances higlight the model’s ability to detect structures that may have been missed or intentially omitted by the human expert.
