Federated AI Networks are Ushering in a New Era in Healthcare Innovation

By Ajit Rajasekharan

Electronic medical record (EMR) adoption has been a primary driver of the increase in healthcare data. While it has led to the creation of enormous volumes of EMR data, most of the data’s potential remains locked at medical centers due to its sensitive nature.

The medical centers holding that data lack the resources or capabilities to maximize its usage for healthcare solutions. Even if they could maximize its usage, analytics platforms often differ between  healthcare systems so the data is only valuable to researchers within that system, limiting the scope and scale of its potential.

In addition, the full potential of EMR data is underleveraged because most biomedical information is unstructured or inaccessible, meaning it cannot be processed or analyzed using conventional data tools or methods.

Several steps are needed to unlock the potential of all that healthcare data. First, the raw data must be de-identified to protect patient privacy. Then the data needs to be harmonized with data from other health systems, broadening the demographic of patients and making it inherently more diverse and equitable. Finally, the data from multiple health systems must be available on a single platform.

The result is a federated AI platform that offers the ability for “one-stop shopping” regardless of the therapeutic area being studied.

With advanced  AI tools and trusted health system partnerships, a federated AI platform can:

• Harmonize structured and unstructured data from across different healthcare systems
• De-identify data, protecting patient privacy
• Detect, extract, and discretize features in unstructured text
• Decipher hidden clues in raw signal waveforms
• Digitize archival and fresh pathology slides for computation
• Synthesize publicly available literature
• Integrate omics data with conventional EMR data
• Recognize artifacts from medical imaging

In this federated AI approach, machine learning models are sent to the location of the data – healthcare centers – where they are locally trained. The models share their learnings via updates with a central server, which houses a global model. The patient-level data never leaves the local location, ensuring  data privacy and meeting regulatory requirements while retaining the benefits of pooled data analysis.

This platform has the ability to unlock a pathologist’s notes about a biopsy or genomic test, a cardiologist’s findings in an echocardiogram or EKG, a radiologist’s observations in an imaging study, a treating physician’s notes about a patient’s symptoms or reactions to treatments, and more.

This multimodal data  provides a richer 360-degree view of the patient, delivering data-driven guidance for decision-making and improving life cycle management. With these insights, researchers can gain a better understanding of patient populations more likely to respond to certain drugs, monitor real-world drug usage, safety, and effectiveness.

A federated AI network can also improve clinical trial design, which is critical because 90% of drugs fail in clinical trials, real-world outcomes are difficult to measure, and clinical trial recruitment is lengthy and costly.

If the trial criteria are poorly defined, it can delay patient recruitment and lengthen the duration of the study. A trial site could close entirely due to a lack of patients. Studies show that 66% of studies incur at least one amendment, adding $500,000 to $1,000,000 in costs and adding six to nine months duration.

A federated AI platform that combines de-identified clinical data from top medical centers with best-in-class AI tools addresses this problem. It can help organizations identify achievable inclusion/exclusion criteria from the beginning, allow them to explore a wealth of multimodal structured and unstructured data, and minimize delays and extra costs often related to inclusion/exclusion amendments.

In addition to supporting  clinical trials, a federated AI platform accelerates discovery and development of new drugs, advances diagnoses and treatments, generates real-world evidence, and most importantly improves patient outcomes.

With the latest large multimodal and language models capable of solving both generative and discriminative tasks in vision, language, and other modalities, a federated AI platform is an appreciating asset to accelerate clinical innovation and release the full potential of healthcare data.

Federated AI is transforming the business of healthcare. For biomedical, device, and pharmaceutical companies not yet leveraging a federated AI platform, now is the time to do so to access data and insights in a usable form – and to find solutions to healthcare’s major problems.

Ajit Rajasekharan is the Chief Technology Officer at nference. He is an accomplished machine-learning practitioner with broad experience in technological innovation.

Prior to joining nference, he was the  vice president of engineering at Rovi, the company that acquired Veveo, which was subsequently acquired by TiVo. Rajasekharan, a co-founder of Veveo, also co-founded Readia, an innovator in digital education.

Prior to Readia he was an engineer at Audible, where he participated in building the world’s first digital audio player and secure digital content delivery system.