Our Datasets
During last years we created a large number of high-quality datasets for many malignant tumors.
Many of these datasets were precisely manually annotated by experts and
are a foundation for our computational pathology algorithms.
If you are interested in cooperation or testing your algorithm on our datasets please feel free to contact us.
Many of these datasets were precisely manually annotated by experts and
are a foundation for our computational pathology algorithms.
If you are interested in cooperation or testing your algorithm on our datasets please feel free to contact us.
#prostate cancer
Prostate cancer: two tumor detection datasets
Two datasets from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Ground truth: patch-level label.
Tissue classes: tumor, benign.
Patch size: 152 x 152 mkm (ca. 600 px)
Magnification: 20x
Dataset 1 (Institute 1): ~145.000 patches
Dataset 2 (Institute 2): ~34.000 patches
Two datasets from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Ground truth: patch-level label.
Tissue classes: tumor, benign.
Patch size: 152 x 152 mkm (ca. 600 px)
Magnification: 20x
Dataset 1 (Institute 1): ~145.000 patches
Dataset 2 (Institute 2): ~34.000 patches
Prostate cancer: Gleason grading dataset
One of the test datasets from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Image source: The Cancer Genome Atlas (TCGA)
Ground truth: Gleason scores, three pathologists.
Image size: large ROIs
Magnification: 20x
Number of images: 218
One of the test datasets from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Image source: The Cancer Genome Atlas (TCGA)
Ground truth: Gleason scores, three pathologists.
Image size: large ROIs
Magnification: 20x
Number of images: 218
Prostate cancer: segmentation dataset
A dataset from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Image source: The Cancer Genome Atlas (TCGA)
Ground truth: precise gland-level annotations.
Tissue classes: 5 tissue classes.
Number of whole-slide images annotated: 389.
A dataset from our study:
"High-accuracy prostate cancer pathology using deep learning published" Nature Machine Intelligence 2020
Image source: The Cancer Genome Atlas (TCGA)
Ground truth: precise gland-level annotations.
Tissue classes: 5 tissue classes.
Number of whole-slide images annotated: 389.
Prostate cancer: six tumor detection datasets
(multiple institutes, multiple scanners)
Six datasets from our study:
"Quality control stress test for deep learning-based diagnostic model in digital pathology"
Modern Pathology 2021
Ground truth: patch-level label.
Tissue classes: tumor, benign glandular, benign non-glandular.
Patch size: 152 x 152 mkm (ca. 600 px)
Magnification: 20x-40x
Size of single datasets: 120 000 patches
(multiple institutes, multiple scanners)
Six datasets from our study:
"Quality control stress test for deep learning-based diagnostic model in digital pathology"
Modern Pathology 2021
Ground truth: patch-level label.
Tissue classes: tumor, benign glandular, benign non-glandular.
Patch size: 152 x 152 mkm (ca. 600 px)
Magnification: 20x-40x
Size of single datasets: 120 000 patches
Prostate cancer: six biopsy datasets
(five institutes, three scanners)
Six datasets from our study:
"An international multi-institutional validation study of deep learning-based classifier for prostate cancer detection and Gleason grading in biopsy samples"
(Under Review)
Ground truth: slide-level labels (tumor/benign/uncertain).
Magnification: 20x-40x
Overall number of whole-slide image: 5922.
(five institutes, three scanners)
Six datasets from our study:
"An international multi-institutional validation study of deep learning-based classifier for prostate cancer detection and Gleason grading in biopsy samples"
(Under Review)
Ground truth: slide-level labels (tumor/benign/uncertain).
Magnification: 20x-40x
Overall number of whole-slide image: 5922.
Prostate cancer: two biopsy Gleason grading datasets
Two datasets from our study:
"An international multi-institutional validation study of deep learning-based classifier for prostate cancer detection and Gleason grading in biopsy samples"
(Under Review)
Ground truth: slide-level Gleason grading (11 pathologists).
Magnification: 40x
Number of whole-slide image: 227 / 159.
Two datasets from our study:
"An international multi-institutional validation study of deep learning-based classifier for prostate cancer detection and Gleason grading in biopsy samples"
(Under Review)
Ground truth: slide-level Gleason grading (11 pathologists).
Magnification: 40x
Number of whole-slide image: 227 / 159.
#oesophageal cancer
Oesophageal cancer: four classification datasets (multiple institutes, multiple scanners)
Four datasets from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Number of institutes: three + TCGA
Ground truth: patch-level (11 tissue classes).
Patch size: 256 px at MPP 0.7813
Size of datasets: 32 796 - 178 187 patches
Four datasets from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Number of institutes: three + TCGA
Ground truth: patch-level (11 tissue classes).
Patch size: 256 px at MPP 0.7813
Size of datasets: 32 796 - 178 187 patches
Link to publication
Oesophageal cancer: four segmentation datasets (multiple institutes, multiple scanners)
Four datasets from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Ground truth: manual annotations.
Magnification: 40x
Number of institutes: three + TCGA
Number of annotated WSIs: 215 / 62 / 214 / 22.
Four datasets from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Ground truth: manual annotations.
Magnification: 40x
Number of institutes: three + TCGA
Number of annotated WSIs: 215 / 62 / 214 / 22.
Oesophageal cancer: one large histological regression grading dataset
One dataset from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Ground truth: case-level, % of vital tumor tissue.
Magnification: 40x
Number of patient cases: 95
Number of WSIs: 1407.
One dataset from our study:
"Artificial intelligence for tumor detection and histological regression grading in oesophageal adenocarcinomas: a retrospective algorithm development and validation study"
Lancet Digital Health 2023
Ground truth: case-level, % of vital tumor tissue.
Magnification: 40x
Number of patient cases: 95
Number of WSIs: 1407.
#colorectal cancer
Colorectal cancer: one large high-quality segmentation dataset
A dataset from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source of cases: TCGA
Ground truth: highly precise manual annotations (17 tissue classes)
Magnification: 40x
A dataset from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source of cases: TCGA
Ground truth: highly precise manual annotations (17 tissue classes)
Magnification: 40x
Number of cases/WSIs: 241
Colorectal cancer: extended CRAG dataset
A dataset from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: CRAG dataset from TIA Center (consider citation of original publications)
Modification: correction of existing annotations, added annotations for nine additional tissue classes.
Ground truth: manual annotations (10 tissue classes)
Magnification: 20x
A dataset from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: CRAG dataset from TIA Center (consider citation of original publications)
Modification: correction of existing annotations, added annotations for nine additional tissue classes.
Ground truth: manual annotations (10 tissue classes)
Magnification: 20x
Number of ROIs: 214.
Link to Zenodo (coming soon)
Colorectal cancer: three segmentation datasets (three institutes, three scanners)
Three datasets from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: three pathology institutes
Ground truth: highly precise manual annotations (11 tissue classes)
Magnification: 40x
Three datasets from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: three pathology institutes
Ground truth: highly precise manual annotations (11 tissue classes)
Magnification: 40x
Number of WSIs: 30 / 10 / 10.
Link to Zenodo (coming soon)
Colorectal cancer: biopsy tumor detection datasets (four institutes, three scanners)
Four datasets from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: four pathology institutes
Ground truth: slide-level labels.
Magnification: 20x-40x
Four datasets from our study:
"Clinical-grade tumor detection and tissue segmentation in colorectal specimens using artificial intelligence tool"
(Under Review)
Source: four pathology institutes
Ground truth: slide-level labels.
Magnification: 20x-40x
Number of WSIs: 356 / 212 / 652 / 675 (total n=1895)
#lung cancer
Lung cancer: large segmentation dataset
A dataset from our ongoing study to lung cancer.
Source: TCGA (adenocarcinoma/squamous cell carcinoma)
Ground truth: highly precise manual annotations (14 tissue classes).
Magnification: 40x
A dataset from our ongoing study to lung cancer.
Source: TCGA (adenocarcinoma/squamous cell carcinoma)
Ground truth: highly precise manual annotations (14 tissue classes).
Magnification: 40x
Number of manually annotated WSIs: 264
(AdenoCa 157 / SqCC 107 )
#kidney cancer
Kidney cancer: large segmentation dataset
A dataset from our ongoing study to kidney cancer.
Source: TCGA + one pathology institute.
Ground truth: precise manual annotations (12 tissue classes)
Magnification: 40x
A dataset from our ongoing study to kidney cancer.
Source: TCGA + one pathology institute.
Ground truth: precise manual annotations (12 tissue classes)
Magnification: 40x
Number of WSIs: 430 (ccRCC 131, pRCC 244, chrRCC 55) + Oncocytomas (n=50).
Kidney cancer: three segmentation datasets
(three institutes)
Three test datasets from our ongoing study to kidney cancer.
Source: three pathology institutes.
Ground truth: precise manual annotations (12 tissue classes)
Magnification: 40x
Tumor entities: ccRCC, pRCC, chrRCC, oncocytomas
(three institutes)
Three test datasets from our ongoing study to kidney cancer.
Source: three pathology institutes.
Ground truth: precise manual annotations (12 tissue classes)
Magnification: 40x
Tumor entities: ccRCC, pRCC, chrRCC, oncocytomas
Number of manually annotated WSIs / institute: 30
#lymph node metastasis
Lung cancer: segmentation dataset
One dataset from ongoing study to lung cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
Tumor entities: AdenoCa, SqCC
One dataset from ongoing study to lung cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
Tumor entities: AdenoCa, SqCC
Number of manually annotated WSIs: 253
Colorectal cancer: segmentation dataset
One dataset from ongoing study to colorectal cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
Tumor entities: Adenocarcinoma
One dataset from ongoing study to colorectal cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
Tumor entities: Adenocarcinoma
Number of manually annotated WSIs: 288
Prostate cancer: segmentation dataset
One dataset from ongoing study to prostate cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
One dataset from ongoing study to prostate cancer metastasis detection.
Source: one pathology institute.
Ground truth: precise manual annotations (6 tissue classes)
Magnification: 40x
Number of manually annotated WSIs: 216
#tissue detection
Tissue detection: segmentation dataset
Source: multiple pathology institutes + TCGA.
Ground truth: precise manual annotations (tissue/background)
Magnification: 40x
Source: multiple pathology institutes + TCGA.
Ground truth: precise manual annotations (tissue/background)
Magnification: 40x
Number of manually annotated WSIs: 208
#histological artifacts
Histological artifact detection: segmentation dataset
Source: multiple pathology institutes + TCGA.
Ground truth: precise manual annotations (normal tissue + 7 histological artifacts)
Magnification: 40x
Source: multiple pathology institutes + TCGA.
Ground truth: precise manual annotations (normal tissue + 7 histological artifacts)
Magnification: 40x
Number of manually annotated WSIs: > 300
© tolklab.de 2023