Inclusive Artificial Intelligence for Accessible Medical Imaging Across Resource-Limited Settings

Projektinformationen

AIMIX

ID Finanzhilfevereinbarung: 101044779

DOI

10.3030/101044779

EK-Unterschriftsdatum 22 September 2022

Startdatum 1 Januar 2023

Enddatum 31 Dezember 2027

Finanziert unter

• European Research Council (ERC)

Gesamtkosten

€ 2 206 963,00

EU-Beitrag

€ 2 206 963,00

Koordiniert durch

UNIVERSITAT DE BARCELONA
Spain

Periodic Reporting for period 1 - AIMIX (Inclusive Artificial Intelligence for Accessible Medical Imaging Across Resource-Limited Settings)

Berichtszeitraum: 2023-01-01 bis 2025-06-30

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

The AIMIX project (Inclusive Artificial Intelligence for Accessible Medical Imaging Across Resource-Limited Settings) is dedicated to advancing AI-driven ultrasound imaging to improve maternal health outcomes in low-resource settings. The project addresses critical gaps in fetal ultrasound accessibility by leveraging federated learning, human-centred AI design, and socio-ethical integration.

In many low-resource environments, access to skilled sonographers is scarce, leading to delayed or suboptimal prenatal care. AIMIX aims to bridge this gap by developing AI-powered solutions that enhance diagnostic capabilities while ensuring usability, affordability, and cultural acceptability. The project's objectives include:
1. Develop the first scientific framework for inclusive AI in medical imaging, and demonstrate its relevance for accessible and effective obstetric ultrasound screening in resource-limited rural settings.
2. Develop integrative-adaptive learning methods that can train inclusive AI algorithms by combining big data from existing repositories with small-size imaging studies from low-resource settings.
3. Develop novel affordable AI methods for accurate estimation of fetoplacental conditions based on robust and deep phenotyping of low-cost fetal ultrasound imaging data.
4. Develop inclusive AI methods and diagnostic tools for obstetric ultrasound imaging that can be used by minimally trained clinicians in rural Africa such as midwives, nurses and technicians.

By integrating AI with local healthcare workflows, AIMIX is expected to contribute to early detection of pregnancy complications, enhance clinical decision-making, and improve maternal and neonatal health outcomes in Africa and beyond. The project’s interdisciplinary approach ensures not only technological innovation but also practical and sustainable implementation in real-world settings.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

Since the project start, AIMIX has made substantial progress across multiple dimensions, including AI model development, data collection, community engagement, and federated learning implementation.

• Data collection and management: Over 1,349 pregnant women were recruited in Kenya, generating a dataset of more than 2,400 standard scans and over 1,900 blind sweeps up to date. This is one of the most comprehensive fetal ultrasound datasets from sub-Saharan Africa, ensuring diversity in AI model training. Additionally, an advanced data de-identification pipeline was developed to comply with data privacy regulations while enabling cross-border AI collaboration.
• AI development: A federated learning-based framework was designed and deployed across five African countries and Spain, significantly improving AI model generalisability without requiring high-end computational resources. This system preserves data privacy while allowing collaborative AI training across multiple sites with different imaging environments.
• Human-centred design and socio-ethical research: A qualitative study with 84 participants highlighted key ethical and cultural concerns regarding AI in maternal healthcare, informing AI model development to enhance trust and usability.
• Technical innovations: The project introduced a novel blind-sweep ultrasound scanning protocol, improving data collection efficiency and AI training. This protocol enables trained healthcare workers with minimal sonography expertise to acquire diagnostic-quality scans, thereby addressing the shortage of specialised personnel.
• Stakeholder engagement: Over 47 community meetings and two major consortium workshops were held, involving healthcare workers, community leaders, and pregnant women to ensure the project's relevance and adoption. AIMIX also established a Community Advisory Board in Kenya to guide implementation and dissemination strategies, ensuring local perspectives are integrated into project decisions.
• Knowledge transfer and capacity building: AIMIX facilitated AI and ultrasound training sessions for healthcare professionals and research personnel, equipping them with skills to sustain and expand the project’s impact. This includes hands-on workshops on federated learning, AI ethics, and ultrasound data interpretation.

These achievements demonstrate AIMIX's effectiveness in translating AI research into real-world applications that can positively impact maternal health outcomes.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

AIMIX has significantly advanced the field of AI-driven fetal ultrasound imaging, pioneering approaches that surpass existing methodologies:

• First federated learning framework for maternal ultrasound in low-resource settings: AIMIX successfully deployed a privacy-preserving federated learning system across diverse clinical environments. This breakthrough enables collaborative AI model training without requiring sensitive patient data to be centralised, a crucial step for AI adoption in global health. The framework’s adaptability ensures that institutions with limited computational power can still benefit from advanced AI insights.
• AI-powered blind sweep protocol for fetal ultrasound: The project developed an AI-assisted blind sweep scanning technique, allowing non-experts to acquire high-quality ultrasound data. This method enhances scalability and accessibility in remote healthcare settings while reducing the dependency on highly trained personnel.
• High-impact socio-ethical AI adoption strategies: AIMIX identified and implemented novel community engagement strategies, including religious leader involvement and culturally tailored AI literacy programs. These insights are now shaping AI deployment frameworks in maternal healthcare.
• Multinational, generalisable AI models: Unlike conventional fetal ultrasound AI models trained on homogenous, high-resource datasets, AIMIX-trained models are optimised for diverse populations, improving robustness and clinical utility. AIMIX’s domain adaptation strategies allow AI models to function effectively across varying imaging conditions and device specifications.
• Knowledge transfer and capacity building: AIMIX trained local healthcare workers, data scientist, and PhD students in AI-driven maternal health technologies, fostering sustainable expertise in resource-limited settings.

By addressing critical gaps in AI usability, fairness, and accessibility, AIMIX has positioned itself at the forefront of AI-powered maternal healthcare solutions, setting new benchmarks for ethical and effective AI deployment in global health. The project’s integration of cutting-edge AI methodologies with real-world clinical applications ensures its sustainability and impact beyond the duration of the grant.

aimix-logo-full-colour-02.png