In support of Breast Cancer Awareness Month, I’m highlighting ways that AI is providing benefits in the prevention, detection, and treatment of breast cancer—and the intellectual property and licensing that supports these advances!
Importantly, the teams generating the groundbreaking research below are sharing code in open source repositories or making the applications commercially available for others’ use and research. The open source model supports rapid evolution of the software as new participants can easily access, and build on, the initial discoveries.
First up—
a large, international study from Google, Northwestern Medicine, and two screening centers in the U.K. published in January 2020 in Nature, uses an AI model that predicts breast cancer in mammography more accurately than radiologists—thereby reducing both false negatives and false positives.
The major components for the code for this project are available in the TensorFlow Model Garden, under the Apache License 2.0. Apache 2.0 is a permissive open source license, meaning it grants full rights to access, modify, and distribute the covered content subject to some basic conditions. A key condition, common to most open source licenses, is referred to as “attribution”—the requirement that any copyright notices or other authorship information be preserved and shared with downstream users. (Apache typically uses a “NOTICE” file for this purpose). Apache 2.0 also provides an express grant of patent rights, subject to the condition (“defensive suspension”) that any rights are terminated if the licensee pursues a patent infringement claim based on the work.
Second—
Mass General Chief of Breast Imaging, Dr. Constance Lehman, worked with colleagues at MIT’s Computer Science and Artificial Intelligence Laboratory to create a model that can analyze subtle patterns in breast tissue that are precursors to breast cancer—patterns the human eye cannot recognize.
The team has shared its code and trained model on GitHub (a popular source code repository). Details can be found the models may be viewed on GitHub (see, e.g.,). According to the documentation, the data may be used for research purposes and supporting tools are available upon request. Commercial use requires a separate license from the author.
Finally—
A team of Swedish researchers found a way to dramatically reduce radiologist workload and allow for earlier detection of critical cases. The researchers accomplished this by using an AI algorithm to triage certain screening mammograms into a no-radiologist work stream. Radiologists then assessed the remainder, and sent certain images into an enhanced assessment.
By freeing up radiologists and using the enhanced assessment stream, this improved process can help more women be diagnosed earlier—and earlier detection is always an improvement. The retrospective study, published in the Lancet in March 2020, used the Lunit INSIGHT MMG AI algorithm, which was developed from a prior study funded by Lunit. Now commercially available from Microsoft AppSource, the Lunit INSIGHT MMG AI solution detects suspicious lesions in mammogram images. Lunit has been granted several patents for its AI inventions.
The licensing for all of this AI research and data is fascinating. This is another example of a permissive open source licenses, that allows others to modify or otherwise build upon the article (provided they accurately credit content from the original authors).
Lunit provided its AI algorithm to the researchers free of charge. The authors used PyTorch for deep learning model development and validation, and made several major components of their work available in the PyTorch open source repository. PyTorch uses a custom open source license providing a broad license grant; the license is largely based on the popular Berkeley Systems Division (BSD) 3-clause license. Datasets included in the OPTIMAM database, which is managed by the St. Luke’s Cancer Centre of Royal Surrey County Hospital in Guildford, England, actually are owned by Cancer Research Technology, and licensed access is granted based on applications made to the OPTIMAM steering committee.
Hats off to these talented scientists and the work they are sharing to improve the health of women around the globe—all with the help of open source repositories and licensing for their cutting-edge discoveries made possible through AI!