Advancing Multi-Institutional Oncology Research with Rhino Health’s Federated Computing Platform

Rhino Health’s Federated Computing Platform (FCP) has transformed how multi-site research consortiums collaborate and share data, offering a comprehensive solution to collaboration and data-sharing challenges. This innovation is vividly demonstrated in a partnership with a consortium funded by the National Cancer Institute, focusing on the critical task of early detection of pancreatic cancer. The consortium, comprising top U.S. cancer research centers, has benefited from Rhino Health’s expertise in Edge Computing and Federated Learning to advance their research efforts to improve pancreatic cancer survival rates significantly.¹

1. Reframing the Approach to Consortium-Based Research

The consortium faced significant delays in traditional data sharing, with agreements taking years and hindering the timely progress of scientific experiments. Rhino Health’s FCP, with its efficient installation process, enabled the consortium to initiate their first collaboration within months. After the one-time installation process, the persistent client can enable additional projects in a matter of days, representing a significant improvement over traditional methods.

Bridging the gap between the logistical challenges of data sharing and the urgent need for progress in medical research, Rhino Health’s FCP emerged as a pivotal solution. By significantly accelerating the collaboration process, the FCP overcame the typical delays in data sharing and set a new standard for efficiency in research collaborations. The advancement in research is particularly crucial when considering the grave and pressing challenges posed by diseases such as pancreatic cancer.

Pancreatic cancer, a formidable adversary in oncology, highlights the urgent need for such advancements. This disease is marked by particularly high morbidity and mortality rates, primarily due to late detection. According to the American Cancer Society, it is estimated² that in 2024, approximately 66,440 people in the U.S. will be diagnosed with pancreatic cancer, and about 51,750 are expected to die from it. The disease accounts for about 3% of all types of cancers and 7% of all cancer deaths in the U.S., underlying the critical importance of rapid and effective research collaboration.

1.1. Importance of Early Detection

Early detection is pivotal. For instance, according to the Surveillance, Epidemiology, and End Results (SEER) database³, maintained by the National Cancer Institute (NCI), which tracks 5-year relative survival rates for pancreatic cancer in the U.S., based on how far the cancer has spread, shows that people who have pancreatic cancer with no signs that the cancer has spread outside of the pancreas have, on average, about 44% of probability of living for at least 5 years after being diagnosed. In contrast, people who have pancreatic cancer spread to distant parts of the body such as the lungs, liver, or bones, have, on average, about 3% of probability of living for at least 5 years after being diagnosed.

1.2. Innovative Approaches to Tackling the Problem

The National Cancer Institute (NCI) funded a consortium network to collaborate with numerous leading U.S. cancer research centers in response to this urgent need. A subset of this network is deeply invested in pioneering early pancreatic cancer detection approaches and understanding their profound impact on patient outcomes.

This network subset is exploring multiple innovative avenues:

• Development of AI Algorithms for Imaging Studies: Using AI to detect early signs of pancreatic cancer in imaging studies, such as CT scans, represents a significant stride in early detection. These algorithms aim to pinpoint subtle disease indications that might be overlooked in standard diagnostic procedures.
• Crowdsourcing Medical Image Annotation: This approach involves enlisting clinical experts to annotate medical images and providing AI models with rich, expert-validated data for training and validation. The challenge lies in the scarcity of such expert annotators, who are often located at remote locations from where the data is located.
• Large Language Models (LLMs) for Clinical Notes Analysis: Another frontier is using LLMs to sift through clinical notes for subtle signs of pancreatic cancer, transforming unstructured data into actionable insights. This technique can unearth early indicators of the disease, which are often buried in routine clinical documentation.

2. Addressing the Core Challenges in Collaborative Research

The central challenge faced by consortiums in medical research, including the one partnered with Rhino Health, lies in the difficulty of data sharing and collaboration. Traditional data sharing agreements can be time-consuming, hindering the timely execution of vital research. Rhino Health’s FCP addresses these challenges head-on, streamlining the process and enabling rapid collaboration. Pancreatic cancer requires extensive and varied data for effective research and development of diagnostic tools. However, no single cancer center possesses a comprehensive dataset to train and refine AI algorithms independently. This need for vast, diverse datasets is critical in developing algorithms that accurately detect early signs of pancreatic cancer across different patient demographics.

2.1. Scalability and Flexibility for Diverse Research Needs

Collaborating with multiple healthcare organizations is essential in compiling a sufficiently diverse dataset. Yet, this collaboration often encounters legal and ethical barriers, particularly regarding patient data sharing. Key issues include:

• Confidentiality and Privacy: Sharing patient data can inadvertently lead to the identification of individuals, raising concerns about privacy breaches and the potential for discrimination or stigmatization.
• Informed Consent: The risk of associating patient data with individual identities without explicit consent poses another significant ethical challenge.
• Safe Handling: Mishandling sensitive patient data can lead to economic and social harm, including identity theft, employment discrimination, and damage to personal and familial relationships.
• Governance Strategies: Adhering to legal and ethical standards for data sharing necessitates robust governance strategies, ensuring sensitive and confidential data is shared in a legally compliant and ethical manner.

Rhino Health’s FCP is highly scalable and features a project-based collaborator model. This allows for the seamless addition of new consortium members and the formation of sub-groups to work on distinct projects. The platform’s versatility facilitates collaborations on various research topics, ranging from imaging studies to Natural Language Processing (NLP) projects, providing a tailored solution for each consortium member’s unique research focus.

2.2. Enhancing Network Growth

The FCP serves the immediate needs of consortium members and opens doors to broader collaborations within the Rhino Health network. This expansive approach enhances the consortium’s capacity for growth and broadens the deployment scope of their research.

To enhance network growth there is a pressing need for a research approach that respects the privacy and security of patient data while enabling collaborative efforts across institutions. This approach must balance data diversity for AI training with the imperative to uphold ethical data handling and sharing standards. A solution that can harmonize these seemingly conflicting needs is beneficial and essential for advancing pancreatic cancer research. This is where innovative solutions like Federated Learning and edge computing come into play, offering a pathway to collaborative research while maintaining the integrity and confidentiality of patient data.

2.3. Powering Grant-Funded Work

Rhino Health’s Federated Computing Platform (FCP) has a proven track record of supporting successful grant submissions and executing grant-funded projects. Large-scale funded projects in the E.U. and U.K. such as the NetZeroAICT consortium⁴, and in the U.S., to name the example presented here, have demonstrated this. This capability is invaluable for consortium leaders looking to secure additional funding to ensure sustainability and attract attention from influential figures in the research community.

2.4. Empowering Research with User Autonomy and Intellectual Property

Rhino Health’s FCP stands out for its user-centric design, serving as a self-service tool that empowers researchers with the flexibility to conduct a wide range of experiments tailored to their unique project demands. This approach is augmented by a subscription-based service, which is carefully crafted to respect the intellectual property rights of researchers to adhere to grant obligations, including mandates like open-sourcing project outcomes. Rhino Health’s commitment to such autonomy and respect for intellectual property is a cornerstone of its role in facilitating cutting-edge, ethical research in the medical field.

3. Rhino Health FCP Innovative Solutions

Rhino Health’s FCP stands out for its scalability and flexibility, essential for multi-site consortiums engaged in various research endeavors. As an illustration, the platform has facilitated collaborations between prominent cancer research institutions on diverse projects. One such collaboration focuses on imaging studies, while another uses the power of NLP (Natural Language Processing) for clinical data analysis. These examples demonstrate the FCP’s adeptness in seamlessly supporting various research needs across different medical fields.

The platform is also designed as a self-service tool, providing researchers with the autonomy to conduct a wide array of experiments tailored to their specific needs. Its subscription-based business model is particularly advantageous, as it does not involve taking intellectual property rights in the projects developed on the platform. This aspect is crucial for researchers looking to fulfill grant obligations, such as open-sourcing their results. By offering this flexibility and respecting the intellectual contributions of researchers, Rhino Health’s FCP facilitates a conducive environment for groundbreaking research while ensuring compliance with academic and funding bodies’ requirements.

The key innovative solutions for the consortium project include:

• Federated Large Language Models (LLMs): The project has pioneered the federation of BioBERT, a cutting-edge model in natural language processing, making a significant milestone in Federated Learning.
• Interactive Containers⁵: Rhino Health’s Interactive Container feature has transformed how medical images are annotated. It enables secure, remote annotation by radiologists worldwide, allowing for rich and expert-validated data essential for training AI models. This feature has been instrumental in palliating the scarcity and high cost of expert medical image annotators.
• Cross-Validation and Algorithm Training: Utilizing the FCP, the project has enabled cross-validation and training of algorithms developed at different cancer centers. This approach allows for detecting pancreatic cancer through AI in imaging studies without exposing the models or the sensitive data, thus maintaining patient privacy and data security.

The outcomes of this project are already making waves in the medical research community. Initial milestones have been presented at the “Bio-IT World, Boston, May 15-18, 2023” conference and published in MIT’s Open Access Article “Federated Learning Framework for NLP in Healthcare: Assessing Hospital Readmission Using Electronic Health Records” by Nalawade, Sahil et al.⁶, with findings indicating significant advancements in hospital readmission predictions using Federated Learning on electronic health records. These results demonstrate the potential of the FCP not only in pancreatic cancer research but across a wide range of medical applications.

3.1. Advanced Capabilities of Rhino Health’s FCP

Rhino Health’s FCP is built with advanced features, ensuring flexibility and scalability to meet the requirements of healthcare data processing as the project progresses:

• Robust Data Encryption and Security: The platform employs state-of-the-art encryption and secure data aggregation, crucial for maintaining data confidentiality.
• Interoperability with Healthcare Systems: It is designed to be highly compatible with various healthcare data formats and systems, facilitating seamless integration into existing healthcare IT infrastructures.
• Sophisticated AI and ML Tools: The platform provides advanced AI model training and validation tools, allowing precise and reliable healthcare AI applications to be developed.
• Edge Computing Technology: Using edge computing, the platform allows data processing on local servers, reducing latency and enhancing data handling efficiency.
• User-Friendly Interface Collaboration: The platform features intuitive interfaces, making it accessible for collaboration across various institutions, enhancing user experience, and facilitating cross-institutional research efforts.

Rhino Health’s FCP is not just a technological innovation but a paradigm shift in how healthcare data is utilized for research, balancing the need for comprehensive data analysis with stringent privacy and security requirements. This platform is setting a new standard in the field of medical research, paving the way for more ethical, secure, and collaborative healthcare studies.

4. Contributions to the Advancements of Healthcare AI

MIT Open Access Article “Federated Learning Framework for NLP in Healthcare: Assessing Hospital Readmission Using Electronic Health Records” by Nalawade, Sahil et al. summarizes the results of a project using Rhino Health’s FCP by concluding that the use of Federated Learning for training large language models on decentralized datasets across multiple hospitals or medical institutions allows for the development of more robust AI models that are less biased and more generalized across diverse patient demographics, while also preserving data privacy and efficiently using limited resources.

Key findings were:

• Enhanced Prediction Accuracy: The study reported increased accuracy for hospital readmission predictions. The Federated Model showed an accuracy of 69.86%, compared to 65.75% with the Centralized Model.
• Improved F-1 Score: The Federated Model achieved an F-1 score of 69.76%, significantly higher than the 64.29% observed in the Centralized Model.
• Increased Sensitivity and Specificity: The sensitivity for the Federated Model was 70.0%, an improvement over the Centralized Model’s 62.07%. Similarly, specificity improved to 70.0% from 69.39%.
• Enhanced Predictive Metrics: The Area Under the Receiver Operating Characteristic (AUROC) and Precision-Recall Curve (AUPRC) also improved. The Federated Model recorded an AUROC of 0.75 and an AUPRC of 0.72, compared to the Centralized Model’s 0.71 and 0.66, respectively.

These results highlight the efficacy of Federated Learning in enhancing the predictive accuracy of healthcare AI models while ensuring data privacy. By enabling collaborative model training across different institutions without direct data sharing, the Rhino Health FCP is instrumental in driving forward the capabilities of AI in healthcare, proving particularly effective in areas where high-quality, diverse datasets are crucial for accurate predictions.

The potential of Rhino Health’s FCP extends beyond pancreatic cancer to other cancer types and diseases. This scalable technology can adapt to various research needs, promising continuous improvement in AI models for broader applications.

5. Setting New Standards in Healthcare AI

Rhino Health’s FCP has facilitated breakthroughs in pancreatic research and provided a model for consortium leaders seeking efficient, scalable, and secure platforms for collaborative research. Its ability to support successful grant submissions and execute grant-funded work underscore its value to consortium leaders looking to engage their research capabilities and secure funding. By integrating innovative technologies like Federated Learning and iGC, the project has shown how collaborative, privacy-preserving research can lead to significant advancements in early cancer detection and overall patient care.

Looking ahead, the project’s approach serves as a beacon of innovation, showing how technology can be leveraged ethically and effectively in medical research. The milestones achieved and the ongoing research promise to transform pancreatic cancer detection and offer insights into other cancer types and diseases.

Experience the transformative power of Federated Learning with Rhino Health. Our Federated Computing Platform innovative features like Federated LLMs and iGC, are ready to elevate your research projects.

Contact Rhino Health today to discover how our Federated Computing Platform can transform your research projects.

‍

‍

^Notes:

^{(1) ITN Imaging Technology News, (2021) EDRN’s Pancreatic Cancer Detection Teams With Rhino Health to Leverage Federated Learning In Order To Accelerate Research, 28 November. Available at: https://www.itnonline.com/content/edrn%E2%80%99s-pancreatic-cancer-detection-teams-rhino-health-leverage-federated-learning-order (Accessed: 25 of January 2024).}

^{(2) American Cancer Society, (2024) Key Statistics for Pancreatic Cancer, 19 January. Available at:  https://www.cancer.org/cancer/types/pancreatic-cancer/about/key-statistics.html (Accessed: 25 January 2024).}

^{(3) American Cancer Society, (2024) Survival Rates for Pancreatic Cancer, 17 January. Available at:  https://www.cancer.org/cancer/types/pancreatic-cancer/detection-diagnosis-staging/survival-rates.html (Accessed: 25 January 2024).}

^{(4) Digital Health, News, Network, Intelligence, (2023). Rhino Health joins consortium to reduce carbon footprint of CT scans. 15 December. Available at: https://www.digitalhealth.net/2023/12/rhino-health-joins-consortium-to-reduce-carbon-footprint-of-ct-scans/ (Accessed: 29 of January 2024).}

^{(5) iGC: Interactive Containers refers to a feature of the Rhino Health Federated Computing Platform. This feature facilitates the remote and secure annotation of medical images by radiologists and medical experts globally. It enables these experts to access and annotate images stored at a single institution from anywhere in the world, enriching the data for AI model training and validation while maintaining data privacy and security. This innovative approach addresses the challenges of expert availability and cost, significantly enhancing the efficiency and accuracy of AI-driven medical research.}

^{(6) Nalawade, Sahil, Samineni, Soujanya, Chowdhury, Alex, Feng, Ling, Umeton, Renato et al. 2023. "FEDERATED LEARNING FRAMEWORK FOR NLP IN HEALTHCARE: ASSESSING HOSPITAL READMISSION USING ELECTRONIC HEALTH RECORDS." Conference Poster at Bio-IT World, Boston, 2023 (May). Available at: https://dspace.mit.edu/handle/1721.1/151725 (Accessed: 25 of January 2024).}

‍