MiDeSeC: A Dataset for Mitosis Detection and Segmentation in Breast Cancer Histopathology Images

2025-07-18

AI Summary

This study addresses the low accuracy of mitosis detection and segmentation in breast cancer histopathological images and the lack of standardized benchmarks. To this end, we constructed a high-quality, domain-specific dataset comprising H&E-stained whole-slide images from 25 patients, acquired at 40× magnification using a 3D Histech scanner and an Olympus microscope. We manually annotated over 500 mitotic instances—including precise bounding boxes and pixel-level contours—within 50 non-overlapping 1024×1024 patches, capturing morphological diversity across cases. The dataset is rigorously partitioned into training and test sets at a 2:1 ratio. It establishes the first unified, reproducible benchmark for mitosis detection and segmentation in histopathology. Empirical evaluation demonstrates that this dataset significantly enhances the robustness and generalizability of deep learning models for mitosis identification and segmentation in pathological imagery, thereby advancing automated cancer grading.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Distributed Search

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

The MiDeSeC dataset is created through H&E stained invasive breast carcinoma, no special type (NST) slides of 25 different patients captured at 40x magnification from the Department of Medical Pathology at Ankara University. The slides have been scanned by 3D Histech Panoramic p250 Flash-3 scanner and Olympus BX50 microscope. As several possible mitosis shapes exist, it is crucial to have a large dataset to cover all the cases. Accordingly, a total of 50 regions is selected from glass slides for 25 patients, each of regions with a size of 1024*1024 pixels. There are more than 500 mitoses in total in these 50 regions. Two-thirds of the regions are reserved for training, the other third for testing.

Problem

• Detecting mitosis in breast cancer histopathology images
• Segmenting mitosis regions for accurate analysis
• Providing a diverse dataset for model training

Innovation

• H&E stained breast carcinoma slides
• 40x magnification scanning
• 1024x1024 pixel regions selection