TRUSTWORTHY AI FOR IMPROVEMENT OF STROKE OUTCOMES

Informazioni relative al progetto

TRUSTroke

ID dell’accordo di sovvenzione: 101080564

DOI 10.3030/101080564

Data della firma CE 11 Aprile 2023

Data di avvio 1 Maggio 2023

Data di completamento 30 Aprile 2027

Finanziato da

• Health

Costo totale € 6 076 437,50

Contributo UE € 6 076 437,50

Coordinato da

FUNDACIO HOSPITAL UNIVERSITARI VALL D'HEBRON - INSTITUT DE RECERCA
Spagna

CORDIS fornisce collegamenti ai risultati finali pubblici e alle pubblicazioni dei progetti ORIZZONTE.

I link ai risultati e alle pubblicazioni dei progetti del 7° PQ, così come i link ad alcuni tipi di risultati specifici come dataset e software, sono recuperati dinamicamente da .OpenAIRE .

Risultati finali

Documenti, relazioni (16)

Locally trained AI models for answering the CEPs

Artificial intelligence trained models with local data from hospitals (VH, PG) for answering each one of the CEPs. The models will be exported for deployment as a REST API, allowing for easy integration and access to the models' predictions The models are trained with the common data model obtained in D3.2. Validation of models is assessed in D4.4. Depends on the final study design: final inclusion criteria, final selection of mandatory and optional variables for retrospective and for prospective cases T6.1 and ethical approval in T6.3.

Design and initial version of TRUSTroke knowledge graph (v1)

The knowledge graph provides TRUSTroke predictions and recommendations with an explainable layer, according to the needs of each use case, and adapted to each target user: doctors and patients for providing interpretable results. It will be defined by the semantically annotated models and data and the results of the trustworthy evaluation assessments. It relates with local trained models D4.3, the validation report D4.4, the explainability score of the models D4.5, should be defined to accomplish security and trustworthiness D4.10, it is the first version of the knowledge graph delivered in D4.7.

Communication and Dissemination plan (initial)

The communication and dissemination plan is a report that will contain fixed protocols, methods and tools for: the creation of a CD kit (logo logo and graphic identity of TRUSTroke); the development of the TRUStroke website, creation of social media accounts; preparation, updating and use of dissemination materials (brochures, online videos and interviews, success stories, flyers, Newsletters, Online/offline publications: blogs, webs, scientific Journals/Magazines local, national, EU levels); organisation of dissemination events.This deliverable will be iterative, with initial deadline at M6 and final at M48. The intermediate versions (due M18 and M36) will be delivered during the corresponding reporting periods.

Project Management Plan

This deliverable will consist of a document that will will define details, tools and mechanisms to ensure efficient management, workflow and work plan, coordination with TRUSTroke internal governing bodies, partners, EC and other stakeholders.

Evaluation Plan

This deliverable will consist of a document that develops a conceptual model of the project, identifies key evaluation points, creates evaluation questions, defines measurable outcomes and develops an appropiate evaluation design. This Evaluation Plan, which includes Quality Assurance and Risks Management, will serve as a reference for partners to carry out continuous monitoring and evaluation, defining the aims and procedures for the evaluation of the results and impact, and to assess KPIs.

Project Management Report (initial)

This deliverable will be produced by the scientific Project Manager hired by the Coordinator (with the funded, requested budget) and will report on the technical and scientific advances of the project, as well as procedures, timing and deadlines for scientific and technical management. It will be updated at M36 and M48.

User-centric framework

This deliverable will describe the initial research conducted by NACAR. They will share and consolidate the learnings within the team through co-creation sessions which will serve as a solid foundation of current knowledge and hypothesis of what current patients’ journey maps are, what users motivations and challenges are and how to potentially improve on them. The conclusions of these sessions will be shared in this deliverable.

Description of user roles, use cases, user journeys

This deliverable is a research report that will contain data, facts and insights that will become the reference for the following development phase. Such information will help define user roles, use cases and user journeys.

Network architectures for first release of the FL network

The deliverable reports on the initial design solutions for the network architecture and the communication protocols (both control and data plane) used to support the first release of the Federated Network platform (R1). Such solutions will be implemented in T2.1 and released in D2.5.

Common Data Model (initial)

This deliverable will contain a report of the designed common data model as well as data harmonisation activities between the different data models present within the hospital partners. The initial version will be prepared in M12 and updated in M24 and M48.

Regulatory plan roadmap for future CE marking and medical use of the TRUSTroke platform

This report aims to indentify applicable technical guidelines to be considered towards the regulatory acceptability of the TRUSTroke platform.

Detailed description of clinical variables

Each clinical centre will draw up a list of clinical and instrumental variables available locally as an expression of daily clinical practice. Potentially useful variables in building predictive models which are transversally available in the 4 clinical centres will be selected. The following data will be considered: clinical history (i.e. gender, history of hypertension, diabetes, dyslipidemia, atrial fibrillation, cardiopathies, previous TIA/ischemic stroke, ongoing antiplatelet/anticoagulant therapy before the index event, timing of symptoms), clinical severity (measured with NIHSS), laboratory exams, neuroimaging findings (occurrence of acute and chronic signs of parenchymal damage, occlusion of intra/extracranial large vessels, collateral circulations), type of acute treatment (thrombolysis/endovascular treatment). This process, coupled with the resources developed in WP 3, will harmonise the existing data resources at the clinical sites and contribute to a Common Data Model (CDM).

FL fundamental algorithms

The deliverable reviews, reports and illustrates the fundamental FL algorithms that will be adopted in the TRUSTroke project. It discusses both relevant machine learning tools and methods to distribute the machine learning tasks over the deployed learners.

Study initiation package

Before enrolment of the first study participant, AI models will be developed and locally trained . Moreover, each institutional ethics committee will have approved the design of the clinical study.

Network security methods and trusted design

The deliverable reports the security and privacy design issues connected to the distributed nature of the FL system, developing a comprehensive threat model and addressing security and privacy considerations and techniques to be implemented in the first release (R1). It also proposes the testing and verification to be performed on the first release (R1) to validate the FL platform.

Open Science management plan

Report concerning open science strategy within the project wich regards outputs publication and repositories.

Piano di gestione dei dati (1)

Data Management Plan (initial)

This deliverable will containt the Data Management Plan of the consortium (the data you acquired and/or generated during the course of the project, management, description, analysis, and storage of suchdata, as well as the mechanisms used at the end of the project to share and preserve the data). The initial version will be delivered at M6, updated at M24 and at M48.

Altro (1)

Workshop sessions

This deliverable will contain the minutes of the Workshop Sessions described in WP1, with the design principles guiding the generation of the low and mid-prototypes of the platform.

Pubblicazioni

Atti di conferenza (5)

Compressed Bayesian Federated Learning for Reliable Passive Radio Sensing in Industrial IoT

Autori: Luca Barbieri, Stefano Savazzi, Monica Nicoli
Pubblicato in: 2024 IEEE Conference on Artificial Intelligence (CAI), 2024
Editore: IEEE
DOI: 10.1109/CAI59869.2024.00071

A Secure and Trustworthy Network Architecture for Federated Learning Healthcare Applications

Autori: Antonio Boiano, Marco Di Gennaro, Luca Barbieri, Michele Carminati, Monica Nicoli, Alessandro Redondi, Usevalad Milasheuski, Sanaz Kianoush, Stefano Savazzi, Albert Sund Aillet, Diogo Reis Santos, Luigi Serio
Pubblicato in: 2024 20th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), 2025
Editore: IEEE
DOI: 10.1109/WiMob61911.2024.10770536

A Carbon Tracking Model for Federated Learning: Impact of Quantization and Sparsification

Autori: Luca Barbieri, Stefano Savazzi, Sanaz Kianoush, Monica Nicoli, Luigi Serio
Pubblicato in: 2023 IEEE 28th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD), 2024
Editore: IEEE
DOI: 10.1109/CAMAD59638.2023.10478391

On the Impact of Data Heterogeneity in Federated Learning Environments with Application to Healthcare Networks

Autori: U. Milasheuski, L. Barbieri, B. Camajori Tedeschini, M. Nicoli, S. Savazzi
Pubblicato in: 2024 IEEE Conference on Artificial Intelligence (CAI), 2024
Editore: IEEE
DOI: 10.1109/CAI59869.2024.00185

First Steps Towards Federated Learning Network Traffic Detection

Autori: Antonio Boiano, Valeria Detomas, Alessandro E. C. Redondi, Matteo Cesana
Pubblicato in: 2024 8th Network Traffic Measurement and Analysis Conference (TMA), 2024
Editore: IEEE
DOI: 10.23919/TMA62044.2024.10559091