AI Healthcare company Presagen has developed a novel Federated Learning technique that can create better performing AI than traditional centralized training approaches.
Federated Learning is a technique that allows AI to train on data distributed in different locations throughout the world, without having to move or centralize the data, in order to protect data privacy. Presagen’s patent-pending approach, called Decentralized AI Training, has AI traveling to the data rather than data traveling to the AI.
Presagen’s Co-Founder and Chief Scientist Dr Jonathan Hall said “With Decentralized AI Training, the AI travels to the data, trains, and then moves to the next data source. Only the AI, which represents general learnings from the data, is shared, and never the private data themselves. This allows our team to train AI on private patient data that we never see.”
An unexpected and exciting result showed that Decentralized AI Training can achieve greater accuracy than training on the same data using the traditional centralized approach. When the data are of poor quality, the technique becomes self-correcting and accuracy improvements of over 10% have been achieved.
Dr Michelle Perugini, Co-Founder and CEO said “This result is critical for developing commercial AI products. Real world problems like healthcare are not Kaggle competitions. Data are inherently poor quality due to subjectivity or clinical uncertainty. Importantly, when you cannot personally see or verify the data, you are exposed to adversarial attacks where data contributors intentionally insert poor quality data. It is important for the AI to handle these situations robustly and automatically for you. This reduces the need to manually process, clean and verify data, which is a huge expense for AI companies, and can also infringe on the privacy of patients.”
Google has its own patented Federated Learning technique originally designed for running AI on individual mobile phones. Presagen’s alternative technique was designed for distributed nodes and clusters of unbalanced data from data sources (e.g. clinics), ensuring no movement or sharing of private data whilst maintaining scalability and efficiency.
Presagen Co-Founder and Chief Strategy Officer Dr Don Perugini said “Federated Learning will become pivotal for AI companies operating in the healthcare industry to build commercially scalable AI medical products. Scalable and unbiased AI need diverse datasets to train on, which represent different patient demographics and clinical settings. However, health privacy laws in many countries prevent private clinical data leaving the country of origin. Federated Learning allows AI to train on a globally diverse dataset, without having to move or centralize the data, to create AI products that clinics and patients anywhere in the world can use and rely on.”
Presagen’s Federated Learning is supported by Presagen’s AI Open Projects initiative, which is currently focused on the underserved women’s health sector (Femtech). Presagen announces a call out for “projects”, which are specific medical products that are being developed, to clinics globally to contribute data. Annotated and verified data from clinics, in a pre-specified format, can be “connected” easily by clinics via the drag and drop function onto Presagen’s online Clinical Data Portal. Data remain on secure local cloud servers within the country of origin, ready for decentralized AI training.
Presagen’s first AI medical product, Life Whisperer, is already in use globally by IVF clinics. Life Whisperer uses AI to identify which embryos in IVF are likely to lead to a successful pregnancy. The assessment is conducted on single static images of embryos. In a published international clinical study Life Whisperer performed 25% better than current manual embryo assessment methods. Life Whisperer was built on globally diverse data from clinics in multiple countries.
Presagen has recently developed a range of patent-pending AI technologies which drives a fundamental paradigm shift in developing commercially scalable AI products for real-world problems, that apply beyond healthcare and to AI more generally.