GoogleGoogle recently partnered with the Stanford University School of Medicine to collect photos of skin diseases across a range of skin colors and body parts to create the "SCIN dataset" for AI training, according to an official press release.The dataset is claimed to "exclusively use photographs submitted by volunteers using the internet" and therefore is claimed to "reflect common skin problems in people".
▲ Image source: Google official press release (the same below)
Google mentioned that many of the industry's medical-specific dermatology image datasets are usually "major diseases", such as people's common rashes, allergies, infections and other photos are usually not in the dataset, so for AI models that should be good at determining people's daily illnesses to use the industry's commonly used medical specialty datasets for training instead of some shortcomings.
Google's SCIN dataset, on the other hand, contains more than 10,000 photos of skin, nail, and hair conditions provided by volunteers and annotated by professional dermatologists, making it more suitable for model testing and training.
The SCIN dataset consists of common allergies, inflammations and infections, with the majority of the images being "early dermatologic traits", with more than half of the dermatologic symptoms appearing within a week of the photo being taken, and 30% of the symptoms appearing in less than a day of the image being taken.Therefore, the relevant dataset can also be used for early detection of skin diseases.