Source-linked AI summary
Lung and Colon Cancer Histopathological Image Dataset (LC25000)
Andrew A. Borkowski, Marilyn M. Bui, L. Brannon Thomas, Catherine P. Wilson, Lauren A. DeLand, Stephen M. Mastorides
TL;DR
Machine-learning pathology research needs large, freely available datasets, particularly for cancer images. This paper creates LC25000, a validated dataset of 25,000 color images in five balanced histologic classes from lung and colon tissue. The dataset is de-identified, HIPAA compliant, and freely downloadable for AI researchers.
Problem
Machine learning requires large datasets, while freely available ML-ready image datasets from cancer pathology and diverse medical entities remain scarce.
Method
The authors created LC25000 by capturing pathology-slide images, cropping them, and expanding them through programmed rotations and flips.
Results
LC25000 contains 25,000 color JPEG images across five classes, with 5,000 images per class.
Takeaways & Limitations
The dataset is de-identified, HIPAA compliant, validated, and freely available for download to AI researchers.
Abstract
from arXiv · showhide
The field of Machine Learning, a subset of Artificial Intelligence, has led to remarkable advancements in many areas, including medicine. Machine Learning algorithms require large datasets to train computer models successfully. Although there are medical image datasets available, more image datasets are needed from a variety of medical entities, especially cancer pathology. Even more scarce are ML-ready image datasets. To address this need, we created an image dataset (LC25000) with 25,000 color images in 5 classes. Each class contains 5,000 images of the following histologic entities: colon adenocarcinoma, benign colonic tissue, lung adenocarcinoma, lung squamous cell carcinoma, and benign lung tissue. All images are de-identified, HIPAA compliant, validated, and freely available for download to AI researchers.
1. Introduction
Machine learning can use labeled pathology images to recognize patterns in new patient images, but successful training requires large, freely available datasets. LC25000 was created to address this need for lung and colon cancer pathology.
- Machine learning can be trained on labeled tissue images to recognize patterns in new patient images and render diagnoses.
- Large image datasets are needed for successful machine-learning training, yet few freely available datasets exist.
- LC25000 provides color images of benign and cancerous lung and colon tissues to address the dataset shortage.
- Lung and colon carcinomas are among the most common invasive cancers and the two most common causes of cancer deaths in America.
2. Dataset
The dataset was built from pathology-slide images, expanded through programmed augmentation, and organized as 25,000 balanced JPEG images across five lung and colon tissue classes.
- Image acquisition: The source material comprised 750 lung-tissue images and 500 colon-tissue images captured from pathology glass slides.
- Image augmentation: Images were cropped from 1024 x 768 pixels to square 768 x 768-pixel images before augmentation.
- Image augmentation: Augmentor expanded the dataset to 25,000 images using rotations up to 25 degrees and horizontal and vertical flips.
- Dataset description: The final dataset contains 25,000 color JPEG images divided equally among five classes of 5,000 images each.
- Dataset description: The files are organized into colon adenocarcinoma, benign colon, lung adenocarcinoma, lung squamous cell carcinoma, and benign lung tissue subfolders.
3. Discussion
LC25000 addresses the need for large, ML-ready cancer pathology datasets by providing 25,000 validated, freely downloadable images across five histologic classes.
- LC25000 contains 25,000 images in five histologic classes, with 5,000 images per class.
- The classes represent colon adenocarcinoma, benign colonic tissue, lung adenocarcinoma, lung squamous cell carcinoma, and benign lung tissue.
- All images are de-identified, HIPAA compliant, validated, and freely available for download by AI researchers.
Funding
The work was supported through resources and facilities at the James A. Haley Veterans’ Hospital.
- The material resulted from work supported with resources at the James A. Haley Veterans’ Hospital.
- The work used facilities at the James A. Haley Veterans’ Hospital.
- The funding statement identifies institutional resource and facility support for the material.