Privacy commissioner to examine I-MED for collecting data to train AI
Make us a preferred source
Make us a preferred source on Google
The news: The privacy commissioner has begun a probe into radiology network I-MED after Crikey reported last week that it had collected data from patients without consent to train artificial intelligence models.
The context: The data was in the form of chest X-ray and CT scans. The AI models in question were not developed by I-MED, but by medtech startup Harrison.ai.
The two companies formed a partnership in 2019 to create Annalise.ai, a platform that can diagnose illnesses by reading chest X-rays and brain CT scans. I-MED provided hundreds of thousands of scans, which were used to train Harrison.ai's models.
Harrison.ai says the data it received from I-MED was de-identifiable — meaning the scans could not be tied to individual patients.
A spokesperson for the Office of the Australian Information Commissioner (OAIC) said it was making "preliminary inquiries" into I-MED.
What they said: "[Companies] must have a clearly expressed and up-to-date policy about their management of personal information, take reasonable steps to notify individuals of how their personal information is used, and can only use or disclose personal information for the primary purpose for which it was collected," they said.
The spokesperson said that data collected could be used for a “reasonably expected” secondary purpose, but that “given the unique characteristics of AI technology, the significant harms that may arise from its use and the level of community concern around the use of AI, in many cases it may be difficult to establish that such a secondary use was within reasonable expectations”.
Harrison.ai co-founder Aengus Tran said: "The data we receive for research and development is de-identified, cannot be re-identified, and is encrypted. Such medical research and development is done in compliance with all relevant regulation, including privacy laws".
I-MED declined to comment.
The sources: OAIC, Harrison.ai
