Bolstering Digital Health Through Federated Learning

Federated Learning

Federated Learning

Federated learning stores training data in the local device and addresses challenges of data privacy.

The novel COVID 19 outbreak has throttled the global healthcare system. As healthcare institutes are ceaselessly working to contain the COVID 19 spread, and treat patients, the treatment and diagnosis of many life-threatening diseases are getting impacted. Through telemedicine consultation, the patients can seek information about healthcare concerns from respective doctors and healthcare specialists. But without physically accessing the health illness in most cases, neither the diagnosis can be formulated nor can the treatment be given.

This brings focus to the role of disruptive technology in healthcare tools and machines. Indeed, artificial intelligence and subsets are getting exponentially adopted in particular domains of healthcare settings. But still many healthcare niches are either devoid or concerned about instituting disruptive technology.

Data created, generated and shared through various healthcare machinery and tools is expounding and prone to security intrusion. Securing the demographic data of the patient cannot ubiquitously eliminate the privacy threat. This is the primary reason why organizations are skeptical about deploying AI-models in the remaining segments of the healthcare infrastructure. Another challenge that limits the disruptive technology adoption is the regulation and sensitiveness of the health data generated. As machine learning models are trained in large datasets, the training of such data is highly regularised in healthcare systems. This implies that even if organizations want to employ AI-models, they are exempted due to policy concerns. Hence organizations are seeking a solution that can deliver positive outcomes while addressing the concerns as mentioned above.


Understanding Federated Learning

Researchers identify federated learning to eliminate privacy and data-sufficiency challenges. In a research paper titled, “The Future of digital health with federated learning,” researchers explore the possibility of providing a solution for future digital health and highlights certain challenges and considerations that need to be addressed.

The traditional machine learning models are governed with a decentralized approach to store the trained data in the cloud platforms. But federated learning stores the training data in the local device. It allows the deployment of smarter models, lower latency and less power consumption while maintaining the individual’s privacy. An example of federated learning is Google’s Gboard in Android phones.


Data-Driven outcomes through a federated approach

Researchers state that federated learning addresses the problem of data governance and privacy by training algorithms collaboratively without exchanging the data itself. It enables gaining insights collaboratively, e.g., in the form of a consensus model, without moving patient data beyond the firewalls of the institutions in which they reside. The machine learning process can occur locally within the institutes. Additionally, the models trained through FL are observed to have high performance compared to the ones trained on centrally hosted data sets.

The researchers also cite that owing to the patient’s privacy concern, each data controller in the FL model, acknowledge the governance processes, privacy policies, and controls data access with the ability to revoke it. This includes both the training, as well as the validation phase. Through this approach, FL could create new opportunities. For example, by allowing large-scale, in-institutional new research can be enabled on rare diseases. Moreover, by moving the model to the data and not vice-versa, it relieves the dual process of duplicating the storage-intense medical data from local institutions in a centralized pool and duplicating it again for every user that uses this data for local model training. It also scales up the potential of growing global data set without necessarily increasing the data storage requirements.



Training such a model will consequentially benefit healthcare infrastructures that are already plagued with privacy concerns. Many consortia like the Trustworthy Federated Data Analytics (TFDA) project and the German Cancer Consortium’s Joint Imaging Platform aim to deploy FL in digital health. However, the intermittent and existing biases associated with algorithms must be addressed while training such a model. Moreover, the manufacturers, clinicians, hospitals, and device handlers need to be highly skilled and well-equipped to benefit from such a model. Researchers observe FL be a promising approach for obtaining powerful, accurate, safe, robust and unbiased models. By enabling multiple parties to train collaboratively without the need to exchange or centralize data sets, FL neatly addresses issues related to the egress of sensitive medical data.

The global federated learning market size is expected to grow from US$1.41 billion in 2017 to US$8.81 billion by 2022, at a CAGR of 44.1%.