Today’s AI is underwhelmingly serving the healthcare industry. Deadlocked by immutable regulations, and rightfully so, safeguarding patient privacy stymies AI developers from applying AI model development with real ground truth patient data. AI has the potential to generate almost $150 billion in estimated potential annual benefits by 2026 to the global healthcare system, but that outcome is largely tied to how efficiently AI developers can access significant amounts of medical data.
According to Mckinsey, 28 percent of healthcare executives reported using AI in their product and/or service development, ranking the third lowest among all industries. Of that fraction, only 8 percent stated that data was accessible by their AI systems. Healthcare data is inoperable and that’s stifling the integration of AI into healthcare, which ironically collects the most data among any industry.
In virtue of the hampering data dynamics, Israeli AI, cybersecurity, data science, and healthcare experts Omer Dror, Ofir Farchy and Dr. Robert Eisdorfer set out to eliminate the barriers constricting medical innovations. They founded Lynx.MD to enable data sharing between hospitals and the community of med-tech and life sciences companies, significantly shortening data access times and the time to market of new digital health technologies. Their cloud-based platform lets AI developers access and study hospital data, and facilitates AI’s integration into healthcare without concern of data privacy and security.
“I was surprised to find how little data was being used by the hospital in the way that decisions are being made on a day to day basis” explained Dror. “A single hospital today doesn’t really have enough data to conduct real deep learning. For instance, a medical center the size of Stanford sees about 30,000 admissions per year, which would mean about 1,000 pneumonia cases - the most common affliction - and subsequently 1,000 chest X-Ray scans with pneumonia per year. These numbers pale in comparison to the Imagenet database released in 2009, which now has over 14 million images, and served as an impetus to the deep learning revolution. Currently, the largest publicly available medical imaging dataset has only 50,000 patients.”
Using cybersecurity best practices and an advanced privacy engine , they ensure complete security and privacy to enable seamless collaboration between AI developers and healthcare providers. The data never changes hands and all AI model training and development happens in the virtual private cloud sandbox of the hospital. To solve concerns for data security during the development engagement, the startup developed a novel distributed federated learning algorithm to allow developers access to hospital’s data without bringing their requested data into the same location, ensuring no private information can be extracted from the hospital’s environment. “It’s an algorithm that doesn't allow for re-identification of data, which is especially critical to the healthcare domain with many naive and edge patient cases” explained Dror.