PathoNet

PathoNet is a general purpose dataset for digital pathology. It consists of 4,462,156 jpg images divided into 12 classes (tissues).

These images were extracted from TCGA (The Cancer Genome Atlas) data portal. No annotations were made, only the tissue type was taken from the slides metadata.

For each tissue, 400,000 256x256 pixel images were randomly selected and downloaded from 400 WSI. An automated cleaning process was then performed to eliminate cases with excessive white content and blurred images.

The dataset is already divided into Train, Test and Validation. When dividing the data, cases were taken into account to avoid mixing images from the same case in different partitions, i.e. all images corresponding to a particular case are in the same partition.

The final number of images for each class and partition are:

Tissue
Partition
# of images

Bladder
Train
308.677

Bladder
Validation
38.927

Bladder
Test
39.166

Brain
Train
313.890

Brain
Validation
39.665

Brain
Test
39.613

Breast
Train
303.949

Breast
Validation
37.499

Breast
Test
38.602

Bronchus and lung
Train
308.848

Bronchus and lung
Validation
37.730

Bronchus and lung
Test
39.160

Colon
Train
243.330

Colon
Validation
30.220

Colon
Test
32.135

Corpus uteri
Train
312.743

Corpus uteri
Validation
39.549

Corpus uteri
Test
39.184

Kidney
Train
311.005

Kidney
Validation
37.950

Kidney
Test
39.184

Liver and intrahepatic bile ducts
Train
314.707

Liver and intrahepatic bile ducts
Validation
38.689

Liver and intrahepatic bile ducts
Test
39.799

Prostate gland
Train
296.181

Prostate gland
Validation
36.568

Prostate gland
Test
36.376

Skin
Train
307.308

Skin
Validation
37.411

Skin
Test
38.487

Stomach
Train
295.002

Stomach
Validation
37.559

Stomach
Test
36.112

Thyroid gland
Train
258.415

Thyroid gland
Validation
33.667

Thyroid gland
Test
33.849

For convenience, the training data has been uploaded by class.