CLASSIC and OE02 Cell Detections
This dataset consists of cell detections, taken from digital whole slide images from two datasts. The method of obtaining these cell detections as well as the sources of the datasets are described in more detail below. For both datasets, the 2D coordinates of the nuclear centroid locations and cell...
Gespeichert in:
Hauptverfasser: | , , |
---|---|
Format: | Dataset |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | This dataset consists of cell detections, taken from digital whole slide images from two datasts. The method of obtaining these cell detections as well as the sources of the datasets are described in more detail below.
For both datasets, the 2D coordinates of the nuclear centroid locations and cell features were extracted using the HeteroGenius MIM Cell-Analysis Add-On (HeteroGenius, Leeds, UK). The model used was a UNet-based cell detector and classifier trained on over 50,000 manually annotated HE-stained cells. 12 nuclear features were extracted. These consisted of length (micrometers), elongation, angle, and the probabiltiy of the clel being one of the following 9 cell types: tumour cell, lymphoycte, granulocyte, plasma cell, fibroblast, smooth muscle cell, endothelial cell, normal epithelium, or other.
One dataset consisted of cell detections from 950 haematoxylin eosin (HE) stained 3mm tissue microarray (TMA) cores from the resection specimen of gastric cancer patients from the CLASSIC trial (Noh et al., 2014). Manual annotations were made of different tissue classes for the purpose of supervised node classification using a graph neural network. These ground truth tissue classes can be found in the column "class" within the csv files. The tissue classes identified were cancer, lymphocyte aggregates, muscle, and stroma. For cells without a class, these were labelled as "notype". Ony 260 TMA cores contained these tissue annotations. A list of these files can be found in the file 'annotated_classic_cores.csv'
The second dataset consisted of cell detections from 45 HE-stained endoscopic biopsies from oesophageal cancer patients from the OE02 trial (Girling et al. 2002). The ground truth target classes in this dataset were "tumour" and "not tumour". Exact annotations of the tumour areas were available from a previous study (Hale et al., 2016) and non-tumour areas were annotated manuyally for a seperate study.
Noh, S. H., Park, S. R., Yang, H.-K., Chung, H. C., Chung, I.-J., Kim, S.-W., Kim, H.-H., Choi, J.-H., Kim, H.-K., Yu, W., Lee, J. I., Shin, D. B., Ji, J., Chen, J.-S., Lim, Y., Ha, S., & Bang, Y.-J. (2014). Adjuvant capecitabine plus oxaliplatin for gastric cancer after D2 gastrectomy (CLASSIC): 5-year follow-up of an open-label, randomised phase 3 trial. The Lancet Oncology, 15(12), 1389–1396. https://doi.org/10.1016/s1470-2045(14)70473-5
Girling, D. J., Bancewicz, J., Clark, P. I., Smith, D. B., Donnelly, R. J., Fayers, P. M., Weeden, S., Gir |
---|---|
DOI: | 10.5281/zenodo.13628225 |