There is a lot of research going on how to use AI in healthcare effectively. It is already being used to improve detection and treatment of disease, discover new medications, identify links between diseases and genes, etc. Virtually any new algorithm has the potential to help patients by analyzing large datasets and finding patterns. AI researchers need access to the correct data to train and test those algorithms.
Hospitals are hesitant to share sensitive patient information with research teams because when the data is shared, they cannot find out researchers are only using the data they need and deleting it after they’re done. Secure AI Labs (SAIL) is taking care of those problems. They have developed a technology that lets AI algorithms run on encrypted datasets. On the owner’s system, data is never left. The usage of datasets can be controlled by healthcare organizations, and the confidentiality of the models and search queries can be protected by the researchers. To collaborate, neither party needs to see the data or the model.
Data from multiple sources can be combined in SAIL’s platform, creating rich insights that fuel more effective algorithms. According to SAIL’s co-founder, “You shouldn’t have to schmooze with hospital executives for five years before you can run your machine learning algorithm. Our goal is to help patients, to help machine learning scientists, and to create new therapeutics. We want new algorithms — the best algorithms — to be applied to the biggest possible data set.”
The company has already partnered with live science companies and hospitals to unlock anonymized data for researchers. According to the company’s co-founder, he found that something is missing in terms of data sharing, whether it was hospitals using old file transfer methods, hard drives, or using emails to send details. In 2017 the idea came to commercialize technology so that AI algorithms run on encrypted data.
Hospitals and other healthcare organizations participating in SAIL’s program make parts of their data available to researchers by setting up a node behind their firewall. Encrypted algorithms are then sent by the company to the servers where the datasets reside. The process is called federated learning. In each server, the algorithms crunch the data locally. The results are then transmitted back to a central model, which updates itself. The data owners, researchers, SAIL all do not have access to the datasets or models.
To further engage the research community at MIT, competitions are being held. This gives access to datasets like protein function and gene expression and challenges researchers to predict results. Machine learning researchers are invited by them to come and train on last year’s data. The researchers have to predict this year’s data.
Suppose it is seen that a new type of algorithm is performing best in these community-level assessments, then it can be adopted locally by people at many different institutions. Researchers can study rare diseases by using this technology. A small pool of relevant patient data is often spread out among many institutions.