MIDOG 2025: Mitotic Figure Detection with Attention-Guided False Positive Correction

2025-08-29

AI Summary

To address the high false-positive rate and poor generalizability in mitosis detection from histopathological images, this paper proposes a two-stage FCOS-based detection framework integrated with a feedback attention mechanism. In the first stage, an enhanced FCOS detector generates initial candidate bounding boxes. In the second stage, a Feedback Attention Ladder CNN (FAL-CNN) models contextual discriminative features, while an attention-guided false-positive correction module jointly optimizes classification and localization, enabling dynamic bounding-box refinement. Feature-level feedback strengthens fine-grained discriminability and effectively suppresses background misclassifications. Evaluated on a preliminary benchmark dataset, the method achieves an F1 score of 0.655—substantially outperforming baseline approaches. This improvement enhances model robustness and clinical applicability for automated mitosis detection in digital pathology.

Technology Category

Computer Vision: Object Detection & CategorizationNatural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)Machine Learning: Calibration & Uncertainty Quantification

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingGraph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphsUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

We present a novel approach which extends the existing Fully Convolutional One-Stage Object Detector (FCOS) for mitotic figure detection. Our composite model adds a Feedback Attention Ladder CNN (FAL-CNN) model for classification of normal versus abnormal mitotic figures, feeding into a fusion network that is trained to generate adjustments to bounding boxes predicted by FCOS. Our network aims to reduce the false positive rate of the FCOS object detector, to improve the accuracy of object detection and enhance the generalisability of the network. Our model achieved an F1 score of 0.655 for mitosis detection on the preliminary evaluation dataset.

Problem

• Reducing false positive rate in mitotic figure detection
• Improving accuracy of FCOS object detector
• Enhancing generalisability of mitosis detection network

Innovation

• FCOS extension with FAL-CNN integration
• Feedback Attention Ladder for false positive correction
• Fusion network adjusts bounding box predictions