A bag of tricks for real-time Mitotic Figure detection

2025-08-27

AI Summary

Mitotic figure (MF) detection in histopathological images suffers from poor robustness and real-time deployability due to inter-center and inter-device variations in scanning hardware, staining protocols, tissue types, and artifacts. To address this, we propose a real-time, cross-center, cross-device MF detection framework. Our key contributions are: (1) a hard-example mining strategy specifically targeting necrotic and fragmented tissue regions; (2) multi-domain balanced sampling coupled with fine-grained data augmentation; and (3) an end-to-end training framework built upon the lightweight RTMDet architecture. Crucially, our method enhances domain generalization and suppresses false positives without increasing inference latency. Experimental results demonstrate strong performance: F1 scores of 0.78–0.84 under five-fold cross-validation across multiple datasets; and on the MIDOG 2025 preliminary test set, the RTMDet-S variant achieves an F1 score of 0.81—outperforming larger models—validating its superior balance of accuracy and computational efficiency.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Mitotic figure (MF) detection in histopathology images is challenging due to large variations in slide scanners, staining protocols, tissue types, and the presence of artifacts. This paper presents a collection of training techniques - a bag of tricks - that enable robust, real-time MF detection across diverse domains. We build on the efficient RTMDet single stage object detector to achieve high inference speed suitable for clinical deployment. Our method addresses scanner variability and tumor heterogeneity via extensive multi-domain training data, balanced sampling, and careful augmentation. Additionally, we employ targeted, hard negative mining on necrotic and debris tissue to reduce false positives. In a grouped 5-fold cross-validation across multiple MF datasets, our model achieves an F1 score between 0.78 and 0.84. On the preliminary test set of the MItosis DOmain Generalization (MIDOG) 2025 challenge, our single-stage RTMDet-S based approach reaches an F1 of 0.81, outperforming larger models and demonstrating adaptability to new, unfamiliar domains. The proposed solution offers a practical trade-off between accuracy and speed, making it attractive for real-world clinical adoption.

Problem

• Detecting mitotic figures across diverse histopathology domains
• Addressing scanner variability and tumor heterogeneity challenges
• Reducing false positives in real-time mitosis detection

Innovation

• RTMDet single stage object detector
• Multi-domain training and balanced sampling
• Hard negative mining on necrotic tissue