Health data management practices in RDI collaboration

• Head: Haoyu Chen, Tenure Track Assistant Professor
• Faculty: Faculty of Information Technology and Electrical Engineering, Center for Machine Vision and Signal Analysis
• Group composition: 5 professors, 2 assistant professors, 5 postdocs, 2 PhD researchers
• Website: https://www.oulu.fi/en/research/human-ai-co-evolution-and-strengthening-human-capabilities/hybrid-intelligence-human-ai-co-evolution-and-learning-multi-realities-hi
• Contact: chen.haoyu (at) oulu.fi

The integration of Artificial Intelligence (AI) into healthcare hinges on the availability of extensive and diverse datasets to develop accurate and reliable predictive models, creating a substantial demand for large-scale data. However, the acquisition and utilization of multi-modal health data are frequently impeded by stringent regulatory restrictions, privacy concerns, and technical barriers. These challenges make cross-institutional and cross-border collaboration in data-driven healthcare innovation particularly complex. To overcome these hurdles, we propose to design a state-of-the-art privacy-preserving federated learning (FL) framework that can train large-scale AI models in a decentralized manner (i.e., without the need of sensitive data sharing between different data controllers) while ensuring stringent data security and superior model efficacy.

The research “Privacy-Preserved Large-Scale AI Learning for Remote Multi-Modal Healthcare with Federated Learning” focuses on building novel privacy-preserving AI frameworks to ensure a trustworthy solution for sharing, training and analyzing large-scale multi-modal health data across different data entities to develop large AI models. The project addresses key challenges in data sharing, decentralized computation, and secure predictive modeling, and aims to apply these methods(but not limited) to the following tasks:

1.Deep Federated Learning Framework for Multi-Modal Data Learning:

Develop decentralized training methods for heterogeneous multi-modal health data (e.g., video, audio, text, and physiological signals) captured across multiple centers. Explore long-term federated training solutions to handle the variability and scale of health data.

2.Privacy-Preserving Large Multimodal AI Model Fine-Tuning:

Build scalable methods for fine-tuning large-scale AI models with federated learning, ensuring privacy protection for sensitive patient data.

3.Remote Mental Health and Emotional Wellbeing Prediction with Decentralized AI models:

Develop AI models for mental health support, leveraging multi-modal data to predict psychological disorders, assess emotional wellbeing, and facilitate remote healthcare delivery.

4.Anonymized Data Sharing and Analysis Practices with Federated Learning:

Incorporate advanced edge computing and anonymization technologies to enable secure data sharing across institutions under federated learning frameworks.

The project is led by Tenure Track Assistant Professor Haoyu Chen, an expert in computer vision and signal analysis, affiliated with the Center for Machine Vision and Signal Analysis (CMVS), at the University of Oulu, in collaboration with Professor Kristina Mikkonen, with the Research Unit of Health Sciences and Technology, Faculty of Medicine, at the University of Oulu , specializing in health technology and ethical healthcare solutions. Their complementary expertise ensures the project addresses both technical and clinical challenges, advancing the intersection of AI and healthcare. The group is dedicated to advancing privacy-preserving AI solutions for healthcare, combining expertise in federated learning, multi-modal data analysis, bioinformatics, and cutting-edge computational methods.

Our group is based on Hybrid Intelligence (HI) research programme at the University of Oulu, funded by the Research Council of Finland in the Profi7 initiative, where we have 5 professors, 2 assistant professors, 5 postdocs, 2 PhD researchers. Through the integration of the key excellence areas in the University of Oulu, HI seeks to combine the strengths of both humans and machines in productive, responsible, and ethical ways.

Please contact chen.haoyu (at) oulu.fi for more details.

Key words: Privacy preserving, healthcare, multimodal AI, large AI models, data sharing, federated Learning

• Head: Kristina Mikkonen, professor & Minna Isomursu, Professor
• Factulty: Faculty of Medicine, Research Unit of Health Sciences and Technology
• Group composition: Healthcare Education and Competence research group, Professor (PI) Kristina Mikkonen, Docent (Professor in UK) Marco Tomietto, Docent Annukka Tuomikoski, Post doc Ashlee Oikarainen, Post doc Miro Koskenranta, Post doc Toni Haapa, Post doc Minna Koskimäki, PhD candidates: Anna-Leena Keinänen (defending January 2024), Sari Pramila-Savukoski (defending March 2024), Erika Jarva (defending April 2024), Suleiman Kamau (defending May 2024), Heli Kuivila (defending June 2024), Jonna Juntunen, Veera Kaarlela, Paula Ropponen, Netta Pohjamies, Virpi Välimaa, Sara Palander. Prof. Minna Isomursu and her research group in 6G enabling sustainable society, Digital Health, data-driven services. Collaborators: POHDE competence development charge nurse scientist Annukka Tuomikoski, BusinessOulu Development Manager, OuluHealth ecosystem
• Websites: https://www.oulu.fi/en/research-groups/health-care-education-and-competence-healtheducom
• https://www.oulu.fi/en/researchers/minna-isomursu
• Contact: kristina.mikkonen (at) oulu.fi

The Wellbeing Services County of North Ostrobothnia (POHDE) encompasses a large area of healthcare staff and services. At the moment, the main operational challenges are the shortage of funding and staff. POHDE has recognised that developing the competences of staff is an essential element of increasing the resiliency and sustainability of healthcare in North Ostrobothnia. New approach to improve the evidence-based healthcare system and enhance healthcare staff competence has been started to ensure the sustainability of the growth of healthcare services in POHDE.

The enhancement of healthcare staff competence is crucial for the reasons of sustaining healthcare organisational growth and ensuring the quality of care for patients. Competent healthcare professionals are better equipped to make accurate diagnoses, implement effective treatment plans, provide appropriate care and prevent medical errors. This directly contributes to improved patient outcomes, including faster recovery times, reduced complications, and overall better health. Also, competence building is necessary for future-proofing the healthcare system to increase resiliency for upcoming challenges.

POHDE has invested in creating systematic capacity to monitor the competences of healthcare professionals. Healthcare professionals’ competence measurement with validated tools (e.g. digital health competence, clinical generic competence, stroke patient care competence) have been integrated into POHDE healthcare staff development practices since 2019 generating large data sets to inform decision-making in staff competence management for leadership. Monitoring is the first step in systematic competence management.

In our project, we will develop artificial intelligence (AI) robust evaluation models of healthcare staff competence by integrating advanced technologies to assess, monitor, and enhance the skills and performance of healthcare professionals. Our process will support the entire process of competence development including 1) data analytics and pattern recognition, 2) performance monitoring, 3) personalised learning paths and hybrid intelligence training.

More precisely, we will use data analytics and pattern recognition by developing AI algorithms, which can analyse vast amounts of healthcare data, including patient outcomes, treatment plans, and staff actions, to identify patterns associated with competent performance. AI-based algorithms will establish benchmarks for competence and provide insights into best practices by correlating data from successful cases with specific actions taken by healthcare professionals. We will also implement AI-driven monitoring systems, which will continuously assess the performance of healthcare staff in real time. This may involve tracking key performance indicators, protocol adherence, and patient outcomes. AI algorithms will identify deviations from expected standards, enabling proactive intervention and targeted training for staff members who may require additional support. As a result personalised learning paths will be developed based on the individual strengths and weaknesses. By analysing historical performance data and identifying specific areas for improvement, AI can recommend targeted educational resources and training modules tailored to each staff member's needs.

The project will be designed considering the requirements of the proposed AI Act and the seven key principles of trustworthy AI systems specified in the AI HLEG ethics guidelines. Special effort will be invested in creating competence building methods tailored for individuals to consider privacy aspects of using person-specific workload data and competence need evaluations.

Key words: Data management and processing, healthcare staff competence management and development, AI evaluation models

• Head: Miika Nieminen, Professor, Chief Physicist
• Co-PI: Simo Saarakkala, Professor
• Faculty: Faculty of Medicine, Research Unit of Health Sciences and Technology
• Group composition: Professors: 2 Postdocs: 3 Researchers: 8
• Website: https://www.oulu.fi/en/research-groups/medical-imaging-diagnostics-algorithms-and-software-midas
• Contact: miika.nieminen (at) oulu.fi

Our research group of Medical Imaging, Diagnostics, Algorithms and Software (MIDAS) is a group working at the interface of the university and university hospital. We are a balanced mixture of university researchers and medical physicists whose core areas of expertise are magnetic resonance imaging (MRI) and computed tomography (CT) imaging. In addition to our academic and hospital connections, we are actively collaborating with various industrial partners.

We have a long history of developing quantitative MRI methods for skeletal diseases, particularly for osteoarthritis and articular cartilage. We focus on developing relaxometry-based methods to identify constituent-specific alterations in tissue ultrastructure, histological correlation of MRI parameters, speeding up MRI image acquisition, developing tools and algorithms for image analysis, and applying these methods in human and patient studies.

CT is one of the most widely used diagnostic imaging techniques. The development of X-ray detectors, computational power, artificial intelligence, and algorithms provides exciting avenues to improve image quality and optimize the radiation dose of the patient. We focus on developing spectral CT, image reconstruction, image enhancement, and artefact suppression methods for CT and cone-beam CT. We collaborate closely with end-users and companies with the aim of bringing the methodologies into clinical use. Through our collaboration with the medical device industry, we have access to the normally proprietary projection data from the scanners. This access allows us to be at the forefront of testing new algorithms with real patient data and diagnostic scanners.

Identification of pathologies from radiological images in one of the most significant boGlenecks in modern diagnostics. Machine learning (ML) based methods have the capacity to automaMze detecMon processes and, hence, reduce human workload. We apply ML to computer vision problems encountered in medical image analysis for multiple modalities and diseases.

As our group works actively with patient data, the new General Data Protection Regulation (GDPR) requires our new projects to utilize a secure operating environment. Developing algorithms and software in such a secure operating environment requires expertise in software development (particularly Python and singularity containers), which is not the core expertise of our group. In this Data4Healthcare call, we are particularly looking for a researcher who wants to work with our focus areas (MRI and CT data processing) and offer their software development expertise to support the group’s migration to such a secure operating environment. As the expertise in using these secure operating environments is limited on a national level, your work would have a significant impact even beyond our group's needs.

Our group is a part of the recently-established Flagship of Advanced Mathematics for Sensing, Imaging and Modelling (FAME), funded by the Research Council of Finland.

Key words: Medical Imaging, Secure Data Environment, Data Processing

• Head: Harri Oinas-Kukkonen, Professor
• Co-PI: Janne Hukkanen, Professor
• Faculty: Faculty of Information Technology and Electrical Engineering, Oulu Advanced Research on Service and Information Systems (OASIS). Faculty of Medicine, Research Unit of Biomedicine and Internal Medicine (INTMED)
• Group composition: OASIS research unit lead by Prof. Oinas-Kukkonen has 2 professors, 6 adjunct professors, 2 postdoc researchers, 15 doctoral researchers. The research group of Hukkanen (INTMED) has 2 professors, 1 postdoc researcher, 4 doctoral researchers as well as 4 study nurses and dieticians and a laboratory nurse
• Websites: https://www.oulu.fi/en/university/faculties-and-units/faculty-information-technology-and-electrical-engineering/oulu-advanced-research-service-and-information-systems
• https://www.oulu.fi/en/researchers/janne-hukkanen
• Contact: harri.oinas-kukkonen (at) oulu.fi

The two research groups, together with an industry partner, jointly form an interdisciplinary and innovative research collaboration for studying novel socio-technical aspects of behavioral medicine, digital health interventions and health behavior change support systems, especially for the prevention and treatment of overweight, obesity, and other lifestyle-associated health conditions.

This domain is critically important as 40% of the determinants of health and their contribution to premature death is due to behavioral patterns – these patterns can be influenced by information technology. The always available modern computing opens up opportunities to match the appropriate digital health interventions with the actual needs of the users. Moreover, these systems maintain powers to motivate users to set goals and encourage and persuade to adopt a desirable target behavior. However, researchers have not really been able to explain what really works, for whom, when, how, and why in these interventions. We study and explain psychological, behavioral and software-related factors associated with successful digital health interventions.

A cornerstone provided for this project is a spin-off from the University of Oulu, known as Onnikka. This persuasive mobile application is provided to the healthcare markets by Onnikka Health company. It enables a vast amount of and ever-growing behavioral, system usage and health outcome data to be analysed parallelly.

Secondary use of data is enabled as data with real patients and participants from three previously conducted large randomized controlled trials (RCT) and one large ongoing European RCT will also be made available as baseline for this project. Health data includes demographic variables, anthropometric measurements such as height, weight, blood pressure, and laboratory values on the most important risk factors of cardiovascular disease and mortality. Crucially, the changes in these variables as affected by the use of Onnikka, in comparison with the appropriate control treatments, are available for the analyses as outcome data.

One of the central ideas in this project is to combine these data with register data obtained from the national health registries such the Finnish Care Register for Social Welfare and Health Care (Hilmo and Avohilmo) and the registry on the medication prescriptions. This register data will enable us to analyze the app usage patterns and individual behavioral traits as the determinants of the health care resource use with an 8-year follow-up (long-term effectiveness and health effects). Another central idea is to search for the predictors of the more immediate effect, i.e., weight change in the RCT data as well as real world data gathered during regular use of the app as a commercially available treatment for overweight and obesity.

Researchers will work at the interface between the academic world and non-academic organisations. Intersectorial exposure is increased by a secondment in the partnering Onnikka Health company. In sum, we bring together a diverse team of experts, including data scientists, information systems designers, and healthcare professionals. §This data science project will develop new methods for data sciences and open up for the applicant an avenue for developing an international scientific career.

Key words: Behavioral medicine, health behavior change, persuasive technology

• Head: Mourad Oussalah, Professor
• Faculty: Faculty of Information Technology and Electrical Engineering
• Group composition: The Center for Machine Vision and Signal Analysis (CMVS) is a creative, open and internationally attractive research center, combining the expertise of University of Oulu computer vision and biosignal analysis scientists. Through expertise spanning decades and extensive international collaboration, CMVS continues to conduct internationally acknowledged work on the cutting edge of its research areas. It includes 7 professors, 11 postdocs and 21 PhD students. The Natural language processing research group includes 2 postdocs and 5 PhD students. Among its activity is digital health where it seeks to mine medical data, including patient electronics health records and social media data, for improving health experience and providing assistence to clinician; developing tools for automatic medical caption generation from medical images; and mining; and comprehending user's activities using body's ubiquitous sensory data and selected set of health indicators.
• Website: https://www.oulu.fi/en/university/faculties-and-units/faculty-information-technology-and-electrical-engineering/center-machine-vision-and-signal-analysis
• Contact: mourad.oussalah (at) oulu.fi

In the era of social media, user's health is dangerously impacted by user's habits, daily activities and social norms that constraint user's actions. For instance, most of currently online / mobile food recommender apps are mainly driven by business insights to promote some restaurant or food service providers, while health aspects are either fully or partially neglected, which contribute to increase in obesity and mobility related diseases and unhealthy eating habits. This research builds on research team track record on social media analytics and food recommender systems that leverage food nutrition content and user’s preference elicited from social media apps or crowdsourcing app to further extend this research to devise mechanisms and frameworks that account for user’s physical and daily activities in a way to guide the user towards healthy diet and develop behavioral interventions for tackling diet / mobility related patterns. For this purpose, this research involves both the development of efficient deep learning models for eliciting user’s preferences and physical / daily activities for generating tailored recommendation, and efficient collaboration with clinicians at University of Oulu, Research Unit of Health Science and Technology and the Finnish Institute for Health and Welfare for developing behavioral based intervention methods. The research expects to contribute to both advance in precision medicine and promote the development of a new business ecosystem centred around healthy diet and increased physical activities. The candidate should have strong experience in machine learning and development of digital health applications. 

Key words: health-recommender systems, physical activity, obesity, machine-learning

• Head: Jarmo Reponen, Professor
• Co-PIs: Arild Faxvaag, Professor, NTNU, Norway; Outi Laatikainen, Post Doctoral Researcher
• Faculty: Faculty of Medicine, Research Unit of Health Sciences and Technology, Faculty of medicine and health sciences, NTNU, Trondheim Norway
• Group composition: Reponen: Professors 3 Postdocs 2 Researchers 3, Faxvaag: 1 professor, 3 PhD students
• Websites: https://www.oulu.fi/en/university/faculties-and-units/faculty-medicine/research-unit-health-sciences-and-technology
• https://www.ntnu.no/inb/helseteknologi
• Contact: jarmo.reponen (at) oulu.fi, arild.faxvaag (at) ntnu.no

Our research group is a multidiciplinary virtual team, which combines the knowledge of the university and the university hospital. Among researchers coming from the Research Unit of Health Sciences and Technology, FinnTelemedicum group PI Prof. Jarmo Reponen has a 20 years history in making research on electronic medical records (EMR) conserning their data content, interoperability and use. Collaborator Prof Miika Nieminen´s group has a long history in developing data analysis and artificial intelligence (AI) methods especially in imaging. Then Co-PI Outi Laatikainen from the Oulu University Hospital (OYS) has nearly 10 years of experience in register studies, focusing on medication safety/pharmacoepidemiology. Adjuct professor Pekka Siirtola and his team from Biomimetics and Intelligent Systems group at the University of Oulu has over 15 years of experience in AI and machine learning.

As healthcare costs have continued to increase, the urgency of reducing low-value treatment and promoting precision medicine has become evident. In lack of continuous real-world evidence (RWE) of e.g. pharmacotherapy effectiveness in unselected cohorts, current treatment guidelines still heavily rely on clinical trials and studies with small cohorts. A major bottleneck in RWE studies worldwide is data extraction, as the current infrastructures do not support such tasks. Therefore we have taken data extraction as our main research theme and made a use case of RWD extraction for a cohort of breast cancer (BC) patients. This approach addresses most of the challenges in terms of variation on health data structures.

Our research platform is the highly appreciated and integrated ESKO EMR system used at OYS, containing longitudinal patient data since late 1990s, the longest time series of all university hospitals in Finland. The in-house company ESKO Systems Oy providing the EMR solution has agreed to facilitate this research initiative.

In this research project, based on earlier efforts, a large and comprehensive data set will be produced, which supports having both BC imaging data and a marked point in time for the progression of the disease. In addition, the cohort is enriched with patient outcome data. In the cohort, there are more than 8,000 first-diagnosis BC patients from northern Finland, with a follow-up period 2005-2022.

Then enriched cohort will be used in the next phase of the project:

1) To further develop an AI-based method where the moment of metastasis can be identified not only based on multimodal analysis of text data but also directly based on the image data and image interpretation.

2) To develop an AI-based method that can be used to evaluate the factors affecting the treatment, and in the future to predict the effectiveness of the treatment in different patient groups/individual patients in advance.

The developed methodologies will benefit not only specific BC treatment decisions but also a wide array of other approaches in all medical fields are possible. Those include assessing what healthcare professionals do with the data and what are the actual outcomes that were achieved (e.g. a learning healthcare systems approach). Another interesting possibility is a digital twin approach, simulation of the effects of treatment decisions.

Key words: Electronic health records; Data extraction; Artificial Intelligence

• Head: Kirsi Sipilä, Professor/ Dental Prosthodontics and Stomatognathic Physiology
• Faculty: Faculty of Medicine, Research Unit of Population Health
• Group composition: Professors: 4 (Kirsi Sipilä, Research Unit of Population Health, University of Oulu, and Oulu University Hospital; Minna Isomursu, Digital Health, University of Oulu; Martti T. Tuomisto (emer.), Clinical Psychology, Faculty of Social Sciences (Psychology), Tampere University; Mikko Tulppo, Department of Physiology, Research Unit of Biomedicine, University of Oulu). Associate Professors: 1 (Ritva Näpänkangas, Research Unit of Population Health University of Oulu, and Oulu University Hospital). Postdocs: 3 (Outi Huhtela, Unit of Dentistry, University of Eastern Finland; Rasmus Valtonen, Department of Physiology, Research Unit of Biomedicine, University of Oulu; Juha Markkula, Turku University Hospital Sleep and Breathing Centre and Psychiatry Outpatient Clinic, and University of Turku). Researcher: 1 (Psychologist Vuokko Hägg, Privat Practice, Espoo)
• Website: https://www.oulu.fi/en/university/faculties-and-units/faculty-medicine/research-unit-population-health
• Contact: kirsi.sipila (at) oulu.fi

Chronic pain affects more than 30 % of population world-wide at some stages of life. Temporomandibular disorders (TMD), including pain conditions linked with masticatory muscles and/or temporomandibular joints, are the second most prevalent musculoskeletal pain condition. Applied relaxation (AR) is a scientifically tested and widely used self-control method for treatment of various psychic and somatic conditions, f ex musculoskeletal pains. The patient is guided to control his/her autonomic reactions and sensations with the aid of relaxation of body. Learning of the method includes several stages using a 6-week program. Digitalization could offer an efficient and flexible use of this method, considering the accessibility, timing, cost-effective use of health care resources, and sustainability. The aim of this randomized, controlled multicenter study is to evaluate the effect of a web application of AR training on TMD pain symptoms, psychosocial symptoms and quality of sleep in TMD pain patients. Additionally, the aim is to study the effect of the AR treatment on autonomic function in a subsample of the TMD pain patients.

The sample includes 124 TMD pain patients from four units of the Student Health Service (YTHS) in Finland. Diagnosis of TMD is based on international, validated Diagnostic Criteria for TMD (DC/TMD). Before treatment, the patients fill in a webropol questionnaire inquiring general health and body pain symptoms, psychosocial symptoms (based on DC/TMD Axis II instruments), and quality of sleep (based on Sleep Condition Indicator). The patients are randomly assigned to two groups (n=62 per group). The splint group is treated as traditionally by a stabilization splint. The AR group uses the web application for training of the method. The questionnaire data are collected before treatment, and 6 weeks, 3 months and 1 year after treatment. Autonomic function (heart rate variability, continuous blood pressure and baroreflex sensitivity) is measured for patients in Oulu YTHS unit (n=31). The patients also use a sleep diary mobile application (Oneiros unipäiväkirja, Oneiros Therapies Oy Ltd) starting 2 weeks before and during the treatment, and 2 weeks before 3 month and 1 year follow-up. The permission of the study has been applied from the Ethical Committee of the Wellbeing Services County of North Ostrobothnia. The duration of the project is 7 years. At present, the web application (Relappy.app) has been made in collaboration with Arcode Oy and is ready for testing. Collaboration is also performed with Oneiros Therapies Oy Ltd.

The present study provides much data from sleep diary registrations and AR web application, questionnaires, and autonomic function. The participants evaluate the state of relaxation before and after each exercise, which also provides data. The analyses of data include comparisons of the variables between the treatment groups at different time points, as well as analyzing the associations between sleep diary parameters and pain variables (intensity of pain, pain-related disability). The topic includes in the theme “Health data management in the practices in RDI collaboration”. The project brings together a diverse team of experts, including data scientists, in development of AR web application as practical solution in healthcare.

Key words: pain, relaxation, web application