Periodic Reporting for period 1 - AI4LUNGS (AI-BASED PERSONALISED CARE FOR RESPIRATORY DISEASE USING MULTI-MODAL DATA IN PATIENT STRATIFICATION)
Berichtszeitraum: 2024-01-01 bis 2025-06-30
AI4LUNGS will develop and validate novel AI-based tools and computational models to improve patient stratification, optimizing diagnosis and treatment of respiratory diseases. The models aim to incorporate clinical partners’ multiple data sources, registries and open national/international databases, including multiple data types from medical records, imaging data as well as novel data from digital stethoscope and –omics. AI4Lungs stratification strategy will build computational models employing structured and unstructured data modalities, leading to more accurate positioning of patients and enabling them to benefit from global data and knowledge shared during all stages of care, focusing on diagnosis and treatment planning. AI and real-world data combined with innovative holistic diseases modelling, will offer a solution for allocating resources more efficiently, making best treatment pipelines accessible to more patients while complying with FAIR principles and relevant regulatory and ethical guidelines. The main objective of the AI4LUNGS project are:
1. Design and build a guideline-based decision support system.
2. Develop a set of integrated and interpretable computational models using AI.
3. Integrate novel data modalities: a) new modality lung auscultation recordings from digital stethoscope examination; and b) new uses for -omics biomarkers.
4. Design, develop and deploy a secure, easy-to-integrate and scalable infrastructure based on industry-accepted protocols and existing systems of KMPG and Yonalink, managing and processing data according to the GDPR, FAIR principles and ethical principles.
5. Design a personalised interactive dashboard.
6. Create an open-access data repository from retrospective and prospective data of approximately 7000 patients (Medical centres).
7. Demonstrate and validate the AI4Lungs tools.
8. Study the impact at all levels of the AI4Lungs concept by performing a Health Technology Assessment (HTA).
9. Maximize impact and design a sustainable exploitation strategy.
10. Analyse and create awareness of ethical, legal and social implications by operationalizing digital ethics principles by applying the Digital Ethical Risk Assessment (DERA®).
AI4LUNGS’ ambition is to successfully create and validate AI tools and computational models to improve patient stratification, ultimately optimising diagnosing and treating respiratory diseases - Interstitial Lung Diseases (ILDs), Infectious Diseases and cancer. It aims to support physicians in their decision-making during the diagnosis and treatment processes, streamlining and digitalising procedures, potentially saving time and reducing the need for numerous examinations compared to traditional approaches. The feasibility and efficiency of the developed solutions will be incorporated into a single system with a UX/UI dashboard to be evaluated in 2 pilots.
1. Design and build a guideline-based decision support system.
2. Develop a set of integrated and interpretable computational models using AI.
3. Integrate novel data modalities: a) new modality lung auscultation recordings from digital stethoscope examination; and b) new uses for -omics biomarkers.
4. Design, develop and deploy a secure, easy-to-integrate and scalable infrastructure based on industry-accepted protocols and existing systems of KMPG and Yonalink, managing and processing data according to the GDPR, FAIR principles and ethical principles.
5. Design a personalised interactive dashboard.
6. Create an open-access data repository from retrospective and prospective data of approximately 7000 patients (Medical centres).
7. Demonstrate and validate the AI4Lungs tools.
8. Study the impact at all levels of the AI4Lungs concept by performing a Health Technology Assessment (HTA).
9. Maximize impact and design a sustainable exploitation strategy.
10. Analyse and create awareness of ethical, legal and social implications by operationalizing digital ethics principles by applying the Digital Ethical Risk Assessment (DERA®).
AI4LUNGS’ ambition is to successfully create and validate AI tools and computational models to improve patient stratification, ultimately optimising diagnosing and treating respiratory diseases - Interstitial Lung Diseases (ILDs), Infectious Diseases and cancer. It aims to support physicians in their decision-making during the diagnosis and treatment processes, streamlining and digitalising procedures, potentially saving time and reducing the need for numerous examinations compared to traditional approaches. The feasibility and efficiency of the developed solutions will be incorporated into a single system with a UX/UI dashboard to be evaluated in 2 pilots.
Work Package 2: End-User Co-Design, System Architecture and Framework Analysis.
• Results from clinical requirements analysis and specification of clinical scenarios (T2.1) mostly available for further R&D work related to technological design (T2.2) medical configuration (T2.3) and user-interaction concepts (T2.4) of cDSS
• First specification results on technical requirements and software-architecture for AI4LUNGS platform (T2.2) available for coordinated software development and integration tasks in related WPs 4, 5, and 6
• Concept for embedding of clinical guidelines into cDSS completed (T2.3) initial configuration of cDSS with digitized relevant guidelines (MS2.2) available for implementation and validation of cDSS prototype in WP 5
• First results for user-AI-interaction concept (T2.4) available for further R&D work related to treatment planning features of cDSS; implementation in interactive dashboard mock-up (MS2.1)
Work Package 3: AI4Lungs Data Repository
• Reach agreement on which data to be collected and used for modelling and analytics (CRFs)
• Request for ethical approval have been submitted and for all except NIPH been approved.
• Test data from all clinical partners have been successfully transferred to YL
Work Package 4: AI Analytic Models for Personalized Diagnosis and Treatment
• Classification methodologies were developed to automatically identify ILD findings in HRCT image slices.
• End-to-end scan classification approaches, using both 2D and 3D CNNs, were implemented for ILD pattern identification, particularly to identify usual interstitial pneumonia (UIP).
• Public X-ray image datasets were used for developing several image finding detection approaches, namely convolutional object detection (YOLO), transform-based object detection (DETR) and vision-language grounded reporting (MAIRA-2).
• Definition of a detailed clinical protocol for digital lung auscultation, applicable to both interstitial lung diseases (ILDs) and infectious diseases (IDs).
• Implementation of baseline deep learning models trained specifically for auscultation data and pre-trained over large sound datasets.
• Definition of the strategic framework for blood-based liquid biopsy data collection, analysis and organization.
Work Package 5: Intelligent Participatory Multimodal AI Decisions for patient stratification
• Draft implementation of multiple correlation analysis methodologies for future integration into T5.1.
• Development of skeleton code structures to support varied multimodal data formats expected from WP3 and WP4.
• Setup of preliminary data preprocessing pipelines in anticipation of real data delivery.
• Early-stage feature selection algorithm designed to interface with the correlation engine.
• Initial architectural and algorithmic exploration for scalable big data analytics in T5.2.
• Definition of the causal inference methodology and a preliminary structure for introspection of the models in T5.3
Work Package 6: AI4Lungs Platform, Digital Twin Service and User Experience:
• definition of the first cloud system prototype implementation strategy
• definition of the update to the cloud system for cost efficiency
• preparation of the 6.1 deliverable to submit on month 19
• provision of DSS in preliminary AI4LUNGS mock-up
• implementation of CRFs in mock-up for documentation of patient cases
• implementation of guideline knowledge in mock-up for case individual treatment recommendations
Work Package 7: Algorithms validation, Pilot Testing & Assessment
• KPI groups defined across core domains: clinical, user experience, technical, economic and policy.
• Partner consultations initiated to refine and validate KPI relevance and applicability.
Work Package 8: Legal and Ethical Issues
• Conducting legal trainings (workshops) on GDPR and AI Act for the consortium.
• Conducting two Digital Ethics Trainings.
• Publishing the AI4LUNGS Digital Ethics Principles.
• Conducting the Digital Ethics Risk Assessment (DERA); Publishing the 1st DERA Report
• Results from clinical requirements analysis and specification of clinical scenarios (T2.1) mostly available for further R&D work related to technological design (T2.2) medical configuration (T2.3) and user-interaction concepts (T2.4) of cDSS
• First specification results on technical requirements and software-architecture for AI4LUNGS platform (T2.2) available for coordinated software development and integration tasks in related WPs 4, 5, and 6
• Concept for embedding of clinical guidelines into cDSS completed (T2.3) initial configuration of cDSS with digitized relevant guidelines (MS2.2) available for implementation and validation of cDSS prototype in WP 5
• First results for user-AI-interaction concept (T2.4) available for further R&D work related to treatment planning features of cDSS; implementation in interactive dashboard mock-up (MS2.1)
Work Package 3: AI4Lungs Data Repository
• Reach agreement on which data to be collected and used for modelling and analytics (CRFs)
• Request for ethical approval have been submitted and for all except NIPH been approved.
• Test data from all clinical partners have been successfully transferred to YL
Work Package 4: AI Analytic Models for Personalized Diagnosis and Treatment
• Classification methodologies were developed to automatically identify ILD findings in HRCT image slices.
• End-to-end scan classification approaches, using both 2D and 3D CNNs, were implemented for ILD pattern identification, particularly to identify usual interstitial pneumonia (UIP).
• Public X-ray image datasets were used for developing several image finding detection approaches, namely convolutional object detection (YOLO), transform-based object detection (DETR) and vision-language grounded reporting (MAIRA-2).
• Definition of a detailed clinical protocol for digital lung auscultation, applicable to both interstitial lung diseases (ILDs) and infectious diseases (IDs).
• Implementation of baseline deep learning models trained specifically for auscultation data and pre-trained over large sound datasets.
• Definition of the strategic framework for blood-based liquid biopsy data collection, analysis and organization.
Work Package 5: Intelligent Participatory Multimodal AI Decisions for patient stratification
• Draft implementation of multiple correlation analysis methodologies for future integration into T5.1.
• Development of skeleton code structures to support varied multimodal data formats expected from WP3 and WP4.
• Setup of preliminary data preprocessing pipelines in anticipation of real data delivery.
• Early-stage feature selection algorithm designed to interface with the correlation engine.
• Initial architectural and algorithmic exploration for scalable big data analytics in T5.2.
• Definition of the causal inference methodology and a preliminary structure for introspection of the models in T5.3
Work Package 6: AI4Lungs Platform, Digital Twin Service and User Experience:
• definition of the first cloud system prototype implementation strategy
• definition of the update to the cloud system for cost efficiency
• preparation of the 6.1 deliverable to submit on month 19
• provision of DSS in preliminary AI4LUNGS mock-up
• implementation of CRFs in mock-up for documentation of patient cases
• implementation of guideline knowledge in mock-up for case individual treatment recommendations
Work Package 7: Algorithms validation, Pilot Testing & Assessment
• KPI groups defined across core domains: clinical, user experience, technical, economic and policy.
• Partner consultations initiated to refine and validate KPI relevance and applicability.
Work Package 8: Legal and Ethical Issues
• Conducting legal trainings (workshops) on GDPR and AI Act for the consortium.
• Conducting two Digital Ethics Trainings.
• Publishing the AI4LUNGS Digital Ethics Principles.
• Conducting the Digital Ethics Risk Assessment (DERA); Publishing the 1st DERA Report
At this stage, AI4LUNGS work in progress of several results that will contribute to the advance of the state of the art, not only on the AI field, but also on the data availability, models interaction and integration with DSS, digital ethics, among others.