Periodic Reporting for period 1 - TRUSTroke (TRUSTWORTHY AI FOR IMPROVEMENT OF STROKE OUTCOMES)
Reporting period: 2023-05-01 to 2024-10-31
Stroke is the leading cause of severe disability worldwide. Stroke is associated with an enormous social and economic burden, which will dramatically increase over the next decades due to population aging. It is of utmost importance for stroke survivors and their families to have access to trustworthy predictions on their probability of recovery to a functional, independent living, as it would greatly contribute to setting appropriate goals for their rehabilitation process and optimizing the patients´engagement on it. Furthermore, the risk of stroke recurrence is high (up to 25%). Tools for a reliable prediction of those patients at a higher risk of recurrence in a given time window would be crucial to prevent early clinical worsening or new ischemic events by means of personalized stroke management.
The main objectives of TRUSTroke project are the following:
Objective 1: A Federated Learning (FL) network to ensure AI robustness and trustworthiness
Objective 2: Trustworthy and validated AI integrated solutions for stroke risk assessment
Objective 3: An app for patient empowerment
Objective 4: A flexible and scalable platform integrating O1, O2 and O3, for continuous upgrading (e.g. with new data types, refined models) and scaling -up (e.g. with new hospitals, new pathologies)
Objective 5: Proof of concept study of TRUSTroke platform in relevant application environment
TRUSTroke proposes advancement versus current state of the art in three main areas of intervention:
1) Creating a trustworthy end-to-end AI solution for better prediction and optimization of patient stroke pathway, reducing the risk and the progression of stroke, integrated into all the phases of the patient care cycle
2) Improving patients quality of life and empowering patients through a better and clearer information about their health situation, answering patient real needs questions
3) Creating a general, scalable, reusable platform to accelerate the development and testing of AI solutions, through the integration of FL technology as well as the interoperability of a chain of different functional blocks which will guarantee the AI solutions to be Trustworthy
TRUSTroke will offer a set of tools empowering the healthcare professionals to take the right decisions to improve the delivery of care and the patient outcomes. The innovation potential in this project lies in the development of novel AI tools trained and deployed on an innovative and robust FL platform that will be fully integrated into the stroke care workflows enabling to dive into the internals of the clinical pathways finding out the real steps performed by patients as well as personalizing risk prediction of early readmission, stroke recurrence and long-term clinical outcome.
For more information: www.trustroke.eu
The main objectives of TRUSTroke project are the following:
Objective 1: A Federated Learning (FL) network to ensure AI robustness and trustworthiness
Objective 2: Trustworthy and validated AI integrated solutions for stroke risk assessment
Objective 3: An app for patient empowerment
Objective 4: A flexible and scalable platform integrating O1, O2 and O3, for continuous upgrading (e.g. with new data types, refined models) and scaling -up (e.g. with new hospitals, new pathologies)
Objective 5: Proof of concept study of TRUSTroke platform in relevant application environment
TRUSTroke proposes advancement versus current state of the art in three main areas of intervention:
1) Creating a trustworthy end-to-end AI solution for better prediction and optimization of patient stroke pathway, reducing the risk and the progression of stroke, integrated into all the phases of the patient care cycle
2) Improving patients quality of life and empowering patients through a better and clearer information about their health situation, answering patient real needs questions
3) Creating a general, scalable, reusable platform to accelerate the development and testing of AI solutions, through the integration of FL technology as well as the interoperability of a chain of different functional blocks which will guarantee the AI solutions to be Trustworthy
TRUSTroke will offer a set of tools empowering the healthcare professionals to take the right decisions to improve the delivery of care and the patient outcomes. The innovation potential in this project lies in the development of novel AI tools trained and deployed on an innovative and robust FL platform that will be fully integrated into the stroke care workflows enabling to dive into the internals of the clinical pathways finding out the real steps performed by patients as well as personalizing risk prediction of early readmission, stroke recurrence and long-term clinical outcome.
For more information: www.trustroke.eu
Work Package 1 (WP1) Patient involvement and user centric approach
Conducted meetings with hospitals and Nora to gather insights into patient experiences and hospital ecosystems.
Developed a patient matrix with five different profiles based on engagement and mobile technology usage.
Conducted workshops to validate findings and created a comprehensive patient journey map outlining phases such as diagnosis, hospitalization, discharge, and post-stroke care.
Designed a visual prototype of the solution ahead of schedule, validated by stroke patients.
Work Package 2 (WP2) Federated infrastructure
Gathered requirements, developed a modular FL platform, and validated its operational functionality.
Defined secure and efficient network architecture using Docker for client nodes.
Designed and optimized FL algorithms, validated with stress tests using public datasets.
Conducted threat and risk assessments for FL infrastructure, addressing vulnerabilities through penetration testing.
Implemented MQTT for FL process management and aligned PaaS functions with IEEE standards.
Work Package 3 (WP3) Data management, trustworthiness and bias
Created a Data Management Plan (DMP) to ensure FAIR and TRUST principles.
Delayed access to VHIR data postponed some tasks, but initial data exploration and Clinical Data Model (CDM) design were completed.
Conducted statistical analyses and developed a Python toolbox for data harmonization.
Prepared AI-ready datasets for stroke analysis and assessed data trustworthiness and bias in line with global ethical standards.
Work Package 4 (WP4) Trustworthy AI models
Established trustworthiness parameters and a prototype repository for data management.
Integrated explainable AI methods (e.g. SHAP, LIME) for stroke outcome prediction, assessing their practicality and effectiveness.
Deployed computational infrastructure for AI model training, addressing delays in server installation at one clinical site.
Developed initial AI models for predictive endpoints, integrated with NORA via a REST API.
Work Package 5 (WP5) TRUSTroke platform integration and patient empowerment
Implemented PROMs and PREMs questionnaires, integrated into a federated software architecture via NORA.
Conducted multilingual translations of the NORA app and ensured secure data routing through a VPN.
Refined the app's design and functionalities, presenting it on AI Appreciation Day for dissemination.
Work Package 6 (WP6) Clinical proof of concept study and contribution to clinical guidelines
Identified clinical parameters for predictive models and finalized retrospective study protocols.
Negotiated data processing agreements and prepared prospective study protocols, including IT infrastructure development for NORA.
Addressed delays in patient recruitment due to ethics committee approvals.
Work Package 7 (WP7) Regulatory acceptability, Exploitation and Sustainability
Established an Ethical Internal Board (EIB) and implemented an Ethics Management Plan.
Created an inventory of intellectual property, analyzing patents and competitive landscapes.
Developed a CE marking roadmap and initiated Health Technology Assessment (HTA) processes.
Work Package 8 (WP8) Project Management, Communication, Dissemination & Open Science
Delivered the Project Management Plan, Evaluation Plan, and Open Science framework on schedule.
Conducted technical reviews, monthly consortium, Executive board, C&D and General Assembly meetings for collaborative alignment.
Launched the project website, social media platforms, and dissemination kits, promoting stakeholder engagement.
Organized webinars to support open science practices and foster knowledge sharing.
Conducted meetings with hospitals and Nora to gather insights into patient experiences and hospital ecosystems.
Developed a patient matrix with five different profiles based on engagement and mobile technology usage.
Conducted workshops to validate findings and created a comprehensive patient journey map outlining phases such as diagnosis, hospitalization, discharge, and post-stroke care.
Designed a visual prototype of the solution ahead of schedule, validated by stroke patients.
Work Package 2 (WP2) Federated infrastructure
Gathered requirements, developed a modular FL platform, and validated its operational functionality.
Defined secure and efficient network architecture using Docker for client nodes.
Designed and optimized FL algorithms, validated with stress tests using public datasets.
Conducted threat and risk assessments for FL infrastructure, addressing vulnerabilities through penetration testing.
Implemented MQTT for FL process management and aligned PaaS functions with IEEE standards.
Work Package 3 (WP3) Data management, trustworthiness and bias
Created a Data Management Plan (DMP) to ensure FAIR and TRUST principles.
Delayed access to VHIR data postponed some tasks, but initial data exploration and Clinical Data Model (CDM) design were completed.
Conducted statistical analyses and developed a Python toolbox for data harmonization.
Prepared AI-ready datasets for stroke analysis and assessed data trustworthiness and bias in line with global ethical standards.
Work Package 4 (WP4) Trustworthy AI models
Established trustworthiness parameters and a prototype repository for data management.
Integrated explainable AI methods (e.g. SHAP, LIME) for stroke outcome prediction, assessing their practicality and effectiveness.
Deployed computational infrastructure for AI model training, addressing delays in server installation at one clinical site.
Developed initial AI models for predictive endpoints, integrated with NORA via a REST API.
Work Package 5 (WP5) TRUSTroke platform integration and patient empowerment
Implemented PROMs and PREMs questionnaires, integrated into a federated software architecture via NORA.
Conducted multilingual translations of the NORA app and ensured secure data routing through a VPN.
Refined the app's design and functionalities, presenting it on AI Appreciation Day for dissemination.
Work Package 6 (WP6) Clinical proof of concept study and contribution to clinical guidelines
Identified clinical parameters for predictive models and finalized retrospective study protocols.
Negotiated data processing agreements and prepared prospective study protocols, including IT infrastructure development for NORA.
Addressed delays in patient recruitment due to ethics committee approvals.
Work Package 7 (WP7) Regulatory acceptability, Exploitation and Sustainability
Established an Ethical Internal Board (EIB) and implemented an Ethics Management Plan.
Created an inventory of intellectual property, analyzing patents and competitive landscapes.
Developed a CE marking roadmap and initiated Health Technology Assessment (HTA) processes.
Work Package 8 (WP8) Project Management, Communication, Dissemination & Open Science
Delivered the Project Management Plan, Evaluation Plan, and Open Science framework on schedule.
Conducted technical reviews, monthly consortium, Executive board, C&D and General Assembly meetings for collaborative alignment.
Launched the project website, social media platforms, and dissemination kits, promoting stakeholder engagement.
Organized webinars to support open science practices and foster knowledge sharing.