TLDR: A new public dataset, NuSeC, has been introduced to help develop AI systems for segmenting nuclei in breast cancer histopathology images. This dataset aims to automate and improve the accuracy and efficiency of breast cancer diagnosis by providing a standardized resource for training and testing computer-aided diagnosis systems.
Breast cancer remains a significant health challenge, being one of the most frequently diagnosed cancers and a leading cause of death among women. Accurate diagnosis is crucial for patient survival, traditionally relying on meticulous examination of tissue biopsies by pathologists. However, this manual process is often time-consuming, labor-intensive, and can be influenced by factors like specimen quality and the pathologist’s experience.
To address these challenges and accelerate the adoption of computer-aided diagnosis (CAD) systems in pathology, a new publicly available dataset called NuSeC has been introduced. This dataset is specifically designed to aid in the segmentation of nuclei within Hematoxylin and Eosin (H&E) stained breast cancer histopathology images. The goal is to foster the development of robust and reliable CAD systems that can streamline and optimize tissue analysis.
The NuSeC dataset was meticulously created using H&E stained breast slides from 25 different patients diagnosed with invasive breast carcinoma no special type (NST). Images were captured at a high magnification of 40x using advanced scanning equipment, including a 3D Histech Panoramic p250 Flash-3 scanner and an Olympus BX50 microscope. A critical step in creating this dataset involved the manual annotation of nuclei structures by experts using QuPath software, ensuring high-quality “mask” images for each original image.
The dataset comprises a total of 100 images, each sized at 1024×1024 pixels. To facilitate consistent research and comparative analysis among future methods, the NuSeC dataset is divided into a training set and a testing set. The training set consists of 75 images, containing approximately 30,000 nuclei structures, while the testing set includes 25 images with around 6,000 nuclei structures. This division ensures that researchers have standardized sets to develop and evaluate their algorithms.
To assess the effectiveness of algorithms developed using NuSeC, two key metrics are recommended: the Aggregated Jaccard Index (AJI) and Intersection over Union (IoU). Both metrics are widely recognized in image segmentation and provide a quantitative measure of how well an algorithm can identify and outline individual nuclei compared to the expert-annotated ground truth. Higher values for these metrics indicate better segmentation performance.
The NuSeC dataset is a valuable resource for researchers and developers working on AI-powered solutions for medical imaging. It is expected to significantly contribute to the advancement of automated breast cancer diagnosis. For more details on the dataset and its creation, you can refer to the full research paper available here.
Also Read:
- Unlocking Detailed Breast MRI Insights with BreastSegNet
- AI Combines Vision Transformers and Graph Networks for Better Breast Cancer Screening
This work was supported by the Turkish Scientific and Research Council (TUBITAK) under Grant No.121E379.
