Company tries image crowdsourcing to speed up AI’s proliferation in radiology

An Australian startup is hoping that it can use global crowdsourcing to accelerate the use of AI in the imaging profession.

Presagen announced Oct. 30 the launch of its AI Open Projects platform, a tool that allows radiology practices worldwide to share images and help to build AI products that are “robust, scalable and unbiased.” Company officials noted that such experimentation can be difficult for small- or medium-sized practices with limited datasets and a lack of resources to pursue machine-learning initiatives.

Oftentimes, AI pilot projects are built around localized data, which can bake biases into the outcomes of these experiments.

"To build AI products that solve global problems, you need a global dataset which is diverse and represents different types of people and clinical settings,” Presagen CEO Michelle Perugini, PhD, said in a prepared statement. “This is challenging because data privacy laws can prevent private medical data leaving the country of origin.  As a result, many focus on building AI from local datasets that are not diverse, creating AI that will be biased and simply will not scale."

Presagen said that it has already put out a call for participants in radiology, along with other specialties such as fertility care. Participation requires data access and clinical support from practices, while it handles AI build, gaining regulatory approval and commercialization. Clinics receive royalties for their participation if the tech hits the market, according to the statement.

The company has already used its crowdsourcing platform to build an AI app called Life Whisperer, which helps providers identify embryos likely to result in pregnancy for couples using in-vitro fertilization. Life Whisperer has performed about 25% better than clinicians using manual methods to predict pregnancy.

Officials noted that Presagen has also perfected a technique that allows its AI to synthesize data without centralizing it in one place. That way, only general knowledge is derived, shared and moved between data sources, and not private patient information.

"Presagen's decentralized AI training technique overcomes significant issues around data sharing to create globally scalable AI. We believe this is a necessary paradigm shift and will revolutionize the way the healthcare industry will build AI products,” Cofounder Don Perugini, PhD, said in the statement.