Sofie Van Hoecke

Sofie Van Hoecke is associate professor at IDLab, Ghent University-imec, and author of more than 200 publications. Her specialties are: data analysis, machine learning, multi-sensor architectures, decision support, semantic dashboards, and the fusion of machine learning and semantic technologies, user feedback and/or physical models, applied in both predictive maintenance and predictive healthcare.

Sofie Van Hoecke is currently leading the PreDiCT research team (http://predict.idlab.ugent.be) of 30 people, funded through Flemish, national and European industry-driven and fundamental projects. As both application domains require active collaborations with actors from industry and healthcare, she developed the skills to work well with people from different backgrounds and to quickly get acquainted with their domain knowledge.

As a researcher and as lead of the PreDiCT research team active in the field of machine learning and hybrid artificial intelligence, her overarching goal is to contribute to the design of innovative solutions to address important challenges in health (both human as well as veterinary health) and maintenance, that have a positive impact on society. She strongly believes in the power of interdisciplinary collaboration and therefore she and her team work closely with experts from other fields, such as human medicine (ICU, neurology, psychology, ...) and veterinary medicine (hip dysplasia, age prediction, CKD cats, …), but also other engineering disciplines, such as chemical and mechanical engineering, to co-design innovative hybrid AI solutions that bridge disciplinary boundaries and drive progress towards shared goals. The team also focusses on bringing code (http://predict.idlab.ugent.be/open_source/) and data (http://predict.idlab.ugent.be/open_data/) into the open source domain. By making code and data available on public repositories, we maximize the scientific impact of our research, facilitate collaboration, and provide a means for other researchers to reproduce and build upon our work in novel ways that may further advance the field.

Domains of expertise

  • Trustworthy AI
  • Hybrid AI
  • Machine learning and decision making
  • Medical decision support
  • Preventive medicine

Selected projects & research platform

  • UGent BOF GOA DEDICAT “Promoting stable remission in depression by combining e-health and cognitive science”

The DEDICAT project aims to develop a follow-up system for psychological functioning. Research shows that there is an increasing need for prevention strategies and early treatment to stay ahead of negative evolutions in mental well-being. At the same time, the surge in technological Innovations and AI has opened new doors to the development of early warning strategies. As people have more and more data about themselves, e.g. through their phones and wearables, we want to use these innovations to do more person-centric monitoring with this data. To do this, we monitor a number of variables from your wearable and phone, related to stress, mood, social interaction and so on, and design hybrid machine learning models on these features to detect early warning signs. We have devised a stepwise research programme towards this is aim with input from psychological science, communication science and engineering, with the input of remitted depressed individuals. As a result, we want to be able to more quickly identify what, for a given person, their specific risk factors or warning signs are, and how these evolve together with their mental health. This research is a collaboration between the PANlab, IDLab, GHEPlab, and mict research groups of UGent.

DEDICAT
  • FWO junior research project HEROI2C “Hybrid Machine Learning for improved infection management in critically ill patients”

Severe infections, common in the ICU, are associated with significant morbidity and mortality. Infection management is challenging here due to uncertainties in antibiotic dosing, and increased antibiotic resistance in this population. Clinicians are today left with inadequate solutions to appropriately dose many of our antibiotics, and have little guidance on who is at risk of nosocomial infections or infections caused by multidrug resistant pathogens, leading to poor outcomes and unacceptable high use of antibiotics, further compromising lifespan of antibiotics and increasing antimicrobial resistance. This project developed hybrid machine learning models to find better solutions for these challenges. We first developed models to predict antibiotic concentrations of the antibiotics used most commonly for severe infections. Secondly, we developed hybrid models for identifying patients at risk of antimicrobial resistant infections. Finally, we made this information available to the healthcare workers at the bedside to tailor the treatment to the patient and the infection, as well as have better insights in the risks the patient is exposed to. This will allow personalized medicine that will improve outcome and reduce antibiotic resistance in these vulnerable patients.

HEROI2C
  • FWO senior research project  “Hybrid AI and Uncertainty Quantification for Trustworthy Time-to-Event Predictions”

In today's healthcare, the prediction of time-to-onset of disease for a particular patient has important diagnostic value; it allows preventive and suitable treatment before symptoms of the disease occur. AI faces challenges in navigating uncertainties inherent in medical data and decision-making, especially in time-to-event predictions as time-to-event data pose challenges due to missingness, e.g. unknown time to event but healthy until last observation. Therefore, this research aims to tackle this challenge by integrating survival analysis data with uncertainty quantification in hybrid AI, ensuring reliable and interpretable time-to-event predictions. Combining these with clinical data and expert knowledge in the hybrid AI models will result in trustworthy solutions to empower the clinicians. Two use cases will exemplify this research. In the first use case, ICU infection management will be provided with dependable predictions for infections of multidrug-resistant organisms and commonly encountered hospital-acquired infections. The second use case will predict insightful prognosis for the manifestation of early chronic kidney disease in cats. However, the methodological advancements fostered by this research project hold the potential to become pioneering, transcending these applications to address a broader spectrum of health issues and even beyond the healthcare domain.https://predict.idlab.ugent.be/projects/

Selected publications

Work details