On Tuesday, the 18th of May, Google unveiled several developments to the public at their annual I/O developer conference. Among them was an AI-powered trial of the "dermatology assist tool", capable of identifying and diagnosing up to 288 known skin conditions.
Reputedly, the AI hopes to provide aid to people who are wary of possible dermatologic conditions, stemming from almost 10 billion total google searches related to skin, hair and nail issues and a significant 2 billion people worldwide actively suffering from such conditions.
Using similar techniques to the ones used by professionals to detect diabetic eye cancer or lung cancer in CT scans, Google's experimental web-based service hopes to provide a list of possible matching conditions a user can suffer from by analysing pictures and asking several questions to narrow down possible ailments, ranging from asking about your skin type to how long you've suffered from the issue. With this data, Google is able to access an extensive and informed database of knowledge centered around all 288 documented conditions and identify which ones users are at risk of having.
Although not intended to be a concrete diagnosis nor a substitute for medical advice, the firm states on their Health blog that they hope to give users access to relevant information and allow them to better inform themselves their next step, may that be attempting home remedies or seeking professional consultation.
Yuan Liu, Ph.D. and technical lead of the project says it took three years of machine learning research and development in order to devise the experimental AI-powered dermatology tool, one of the factors being, according to the technical lead, accounting for health disparities creating biased data.
The firm emphasises the thorough testing that their AI went through in order to identify and accommodate for a range of external factors - age, gender, race and skin type included - in order to narrow the margin of error. Reportedly, over 65,000 images of diagnosed skin conditions and millions for examples of healthy skin were used to educate the AI, all across different demographics.
The AI model used to power the dermatology tool recently passed clinical validation tests, and has been CE marked as a Class I medical device in the EU in line with European health and safety standards, however it is yet to be evaluated by the Food and Drug Administration (FDA) for use in the USA.
Comments