Breast cancer poses a significant global health challenge, with artificial intelligence (AI) seen as a pivotal tool for enhancing diagnostic techniques. Despite AI’s potential, accessing diverse and geographically scattered data remains a major hurdle due to stringent privacy laws. Rhino Health’s Federated Computing Platform (Rhino FCP) addresses these challenges by integrating a privacy-preserving, collaborative approach to data analysis. This innovation is showcased through our partnerships with Assuta Medical Center, Emory University, and I-Medata AI Center in Tel Aviv Sourasky Medical Center (Ichilov) and funded by ELISE (European Network of AI Excellence Centres). These collaborations use the Rhino FCP Federated Computing and Federated Datasets to refine breast cancer diagnostic models, highlighting its crucial role in facilitating secure, efficient data utilization across borders and enhancing patient outcomes.

“This project and Rhino's Federated Computing capabilities enable me to take my research to the next level, as I have been able to validate and improve the generalizability of algorithms by securely leveraging diverse data from across the world, all while maximizing privacy.”

—German Corredor Prada, PhD, MS, Assistant Professor from Emory University

Global Urgency of Breast Cancer Research

Breast cancer is the second most diagnosed cancer globally and a leading cause of cancer death. In 2022, it accounted for 2,296,840 new cases¹; nearly 685,000 deaths were recorded in 2020². By 2040, numbers are projected to increase to over 3 million new cases and 1 million deaths annually, driven by demographic changes like aging and population growth². These stark figures underline the urgent need for enhanced diagnostic methods. This collaborative project effectively addresses these needs using Federated Learning and Edge Computing. The advancements made through this project demonstrate the capabilities of AI in improving diagnostic accuracy and showcase how Federated Computing can bridge the gap between data accessibility and stringent privacy regulations, thereby directly impacting patient outcomes on a global scale.

Innovative Data Handling Approach

Rhino Health’s Federated Computing approach uses Federated Learning and Edge Computing and is characterized by decentralized data processing and decision-making. This ensures sensitive data remains at its source or original location, making it possible to perform complex computations and analyses securely at the ‘edge’ of the network—near the data’s location where it has been stored. This approach maintains strict data privacy and enhances computational efficiency.

‍

Cross-Atlantic Collaborative Efforts

The collaboration among Rhino Health and Emory University in the USA, and Assuta Medical Center and I-Medata AI Center in Tel Aviv Sourasky Medical Center (Ichilov) in Israel, exemplifies a practical application of the Rhino Health’s Federated Computing approach:

1. Algorithm Development and Training at Emory University: Development and preliminary training of a deep learning algorithm to detect mitotic features in pathology images using the publicly available MIGDOG++ dataset.
2. Cox Proportional Hazards Model Training: Employing a Cox proportional hazards model with elastic net penalty, using the TCGA-BRCA dataset distributed across multiple nodes.
3. Federated Validation at Assuta Medical Center and I-Medata AI Center in Tel Aviv Sourasky Medical Center (Ichilov) : Real-world validation of the algorithm used a dataset of 500 patient records from Assuta and I-Medata AI Center in Tel Aviv Sourasky Medical Center (Ichilov), testing its clinical efficacy within the medical center’s network.

The international effort highlights the project’s  global impact and showcases Federated Computing’s efficiency in bridging geographical and institutional divides, thereby accelerating the pace of medical research and innovation across continents within six months.

‍

‍

Benchmark Achievements

The collaborative project led to new benchmarks in breast cancer diagnostic  accuracy. Initially, Emory University used the publicly available MIDOG++ dataset for algorithm training. This phase primarily enhanced the model’s capability to recognize diverse pathological features across various tumor types. The project’s success was quantified by achieving a mean average precision of 0.82 at a threshold of 0.5 on the MIDOG++ dataset, establishing a new benchmark for mitosis detection.

While this high level of accuracy highlights the Rhino FCP capability  to support Federated Learning, the true innovation lies in applying these findings to Federated data, thereby demonstrating the algorithm’s generalizability. This was achieved when Assuta Medical Center and I-Medata AI Center in Tel Aviv Sourasky Medical Center - Ichilov  curated a unique dataset that included 500 annotated breast cancer cases. The novelty of such an approach is highlighted by the incorporation of Edge Computing, where the data remained securely behind the medical center’s local firewall. This setup allowed Emory University researchers  to validate their diagnostic model using the data, without it ever leaving the hospital’s premises, ensuring maximum data security and compliance with privacy regulations. 

Looking Ahead

With the rising global incidence of breast cancer², there is an urgent need for advanced diagnostic and therapeutic strategies. The project significantly advanced AI-assisted diagnostic prediction by eliminating barriers to data sharing.

We invite the medical and research community to explore our Federated Datasets and other resources available through our Federated Computing Platform (Rhino FCP). These Federated Datasets offer unparalleled opportunities for developing and testing new medical solutions securely and efficiently.

‍

To learn more about our Federated Datasets, please get in touch with us today!

‍