Periodic Reporting for period 1 - FLUTE (Federate Learning and mUlti-party computation Techniques for prostatE cancer)
Reporting period: 2023-05-01 to 2024-10-31
In an era where healthcare data is exponentially growing, the FLUTE project emerges as a trailblazing initiative aimed at harnessing this data for medical advancements while ensuring data privacy. Located at the intersection of healthcare, artificial intelligence (AI), and data security, FLUTE aims to elevate and scale up data-driven healthcare on an international level.
One of the cornerstone issues the project tackles is privacy. In the globalized age, cross-border data sharing is crucial for healthcare advancements. However, concerns around data security and privacy of sensitive data such as health data are significant roadblocks. FLUTE aims to resolve this by pioneering a secure multi-party computation within Federated Learning, making it possible for various data hubs to collaborate without compromising on data privacy. This innovative approach aims not only to facilitate collaboration but also to ensure GDPR compliance, aligning with the European Commission's directives.
To demonstrate the practical impact of such platform, the FLUTE platform focuses its first application on clinically significant prostate cancer. By integrating the FLUTE platform with health data hubs from three different countries, a novel federated AI toolset will be developed and validated multi-nationally with the objective to improve the diagnosis of aggressive forms of prostate cancer
Multidisciplinary by nature, the project incorporates social sciences and ethics, reflected in the composition of the 12-strong consortium that includes clinical experts, technology SMEs, research partners, legal/ethical advisors, and a standards organization. The project also promises significant contributions to the global HL7 FHIR standard development, an essential protocol for global healthcare data exchange.
One of the cornerstone issues the project tackles is privacy. In the globalized age, cross-border data sharing is crucial for healthcare advancements. However, concerns around data security and privacy of sensitive data such as health data are significant roadblocks. FLUTE aims to resolve this by pioneering a secure multi-party computation within Federated Learning, making it possible for various data hubs to collaborate without compromising on data privacy. This innovative approach aims not only to facilitate collaboration but also to ensure GDPR compliance, aligning with the European Commission's directives.
To demonstrate the practical impact of such platform, the FLUTE platform focuses its first application on clinically significant prostate cancer. By integrating the FLUTE platform with health data hubs from three different countries, a novel federated AI toolset will be developed and validated multi-nationally with the objective to improve the diagnosis of aggressive forms of prostate cancer
Multidisciplinary by nature, the project incorporates social sciences and ethics, reflected in the composition of the 12-strong consortium that includes clinical experts, technology SMEs, research partners, legal/ethical advisors, and a standards organization. The project also promises significant contributions to the global HL7 FHIR standard development, an essential protocol for global healthcare data exchange.
In a first phase, WP1 studied the requirements and several other WP studied the state of the art.
In month 12 a new partner, SR, with a profile relevant to the studied requirements, was added through the hop-on facility, together with a new WP9.
Next, the use case was prepared (including ethical committee approval, regulatory compliance, and data acquisition, preprocessing and integration) while WP2, WP3, WP4 and more recently WP9 started both the planned research and initial development of components for the FLUTE platform.
Halfway the project, we are now integrating the first version of all components (platform, PET strategies, AI algorithms, dashboards, data) so we can test the cooperation of the results of the several project parts.
In month 12 a new partner, SR, with a profile relevant to the studied requirements, was added through the hop-on facility, together with a new WP9.
Next, the use case was prepared (including ethical committee approval, regulatory compliance, and data acquisition, preprocessing and integration) while WP2, WP3, WP4 and more recently WP9 started both the planned research and initial development of components for the FLUTE platform.
Halfway the project, we are now integrating the first version of all components (platform, PET strategies, AI algorithms, dashboards, data) so we can test the cooperation of the results of the several project parts.
Progress was made towards multiple results beyond the state of the art.
In WP2, we are contributing to scalability of privacy-preserving federated learning by (a) developing optimization algorithms with smaller communication cost, (b) developing multi-party computation algorithms suitable for large-scale sparse data, and (c) integrating trusted execution environments into the FLUTE platform.
In WP3 we are developing new synthetic data algorithms, we are generalizing them towards federated algorithms and we fit existing and new techniques to the prostate cancer use case.
In WP4, we are working on the FLUTE platform for scalable privacy-preserving learning with a set of features well beyond the TRUMPET platform of which it will reuse key components.
WP5 and WP6 are making contributions to the use case (prediction of prostate cancer) and towards standardization.
These results will lead in the second half of the project to impact as described in the dissemination plan, e.g. for several of the advances in WP2 and WP3 we are preparing publications for peer review. In the longer run, to maximize uptake it will be important to align well our technical results and societal and regulatory expectations (e.g. as laid out in the AI Act).
In WP2, we are contributing to scalability of privacy-preserving federated learning by (a) developing optimization algorithms with smaller communication cost, (b) developing multi-party computation algorithms suitable for large-scale sparse data, and (c) integrating trusted execution environments into the FLUTE platform.
In WP3 we are developing new synthetic data algorithms, we are generalizing them towards federated algorithms and we fit existing and new techniques to the prostate cancer use case.
In WP4, we are working on the FLUTE platform for scalable privacy-preserving learning with a set of features well beyond the TRUMPET platform of which it will reuse key components.
WP5 and WP6 are making contributions to the use case (prediction of prostate cancer) and towards standardization.
These results will lead in the second half of the project to impact as described in the dissemination plan, e.g. for several of the advances in WP2 and WP3 we are preparing publications for peer review. In the longer run, to maximize uptake it will be important to align well our technical results and societal and regulatory expectations (e.g. as laid out in the AI Act).