Collaborative AI for Oncology Research

at Dana Farber

Oncology Research Labs are at the forefront of cancer classification research. Facing the challenge of analyzing vast oncology data spread across multiple institutions, they often seek a solution that allows for model training on large datasets without the risk of exposing sensitive data.

The problem

Challenges in Collaborative Research

Accessing and utilizing large-scale oncology data from multiple medical centers introduces significant privacy and security challenges. Pathology images, considered Protected Health Information (PHI), cannot be freely shared among organizations, necessitating a secure method of collaborative analysis.

The Solution

Medical Center Data Remains Secure

Duality Technologies, has developed a secure collaboration platform with federated learning and trusted execution environment that adheres to stringent security and privacy standards. This framework maintains a structure where each medical center’s data stays securely on-site. The model training occurs locally, and only the model weights are transmitted. These weights are aggregated on a global server, and to ensure maximum security, Duality employs a trusted execution environment. This guarantees that the weights are encrypted during transmission and remain inaccessible to unauthorized parties.

Usability

The platform is designed for ease of use without requiring extensive technical knowledge from participants.

Improved Accuracy

Improved model accuracy by training model on images from multiple medical institution.

High Security

Advanced encryption techniques prevent potential adversaries from intercepting or recreating the model from network traffic.

Flexible and Scalable

The platform supports various models and scales according to the collaborative needs.

Privacy Ensured

No direct data sharing occurs between institutions.

Cloud Agnostic

Easily deploy in any environment (AWS, GCP, Azure) to minimize data movement while maximizing data security and privacy.

DANA FARBER CASE STUDY

Secure Federated Learning for Oncology Research

The secure federated learning study by the Dana Farber Cancer Institute in collaboration with Duality Technologies showcases significant operational benefits for medical research, including agility, flexibility, and privacy-protection for sensitive data for all participants in federated collaborations.

This approach demonstrates full control over sensitive medical data by the data custodians, eliminating the need for data transfer, as well as many additional benefits, proving to be a pivotal solution for multi-institutional collaborative oncology research.