Active Learning for Cooperative Optimization of Predictors for intensive care units

Short description

Intensive care units (ICUs) are an essential part of the care of many patients with severe acute conditions. As such, ICUs are a bottleneck that was particularly evident during the Covid crisis. Intensivists recognize the need for AI to optimize their specialty, but it is still an under-explored area. Yet it is the medical specialty in which the most data is collected.

The objective of the ALCOP project is to develop decision support systems for ICU staff in various clinical areas such as hemodynamics and vascular, respiratory and infectious pathophysiology. ALCOP will also focus on the prediction of morbidity and mortality as well as the development of new innovative scores and indicators useful for diagnosis, therapy and prognosis to improve the management of individual patients. More globally, ALCOP will address the issue of optimization and organization of intensive care services based on predictions of length of stay, human resources, hospital capacity, equipment and medical procedures. The project will implement various machine learning technologies including a distributed learning infrastructure via learning coalitions to address the multi-centric dimension, transformers-based learning models to address the heterogeneity and multimodality of data from intensive care units. It will also focus on the interpretability by medical staff of the decisional models obtained, which will ensure their real use in the field.

The results of the ALCOP project will be valorized by the company Eonix in the form of a platform for predicting the needs of intensive care units for the optimization of clinical and organizational decisions.

Time needed : 36 months


1 (University) 

Université catholique de Louvain (UCLouvain)  


2 (University) 

Université Libre de Bruxelles (ULB) 


3 (University) 

Université de Mons (UMons) 


4 (Approved research center) 



Status of the project

Submitted to the Win2Wal Call on March 1, 2022: