Ikerlan – Federated Models for Predicting the Evolution of ICU Patients
Ikerlan
Sector: Health
Business Case
The company currently aims to deliver technological solutions for the healthcare sector, focusing on two main aspects: first, developing predictive models for assessing ICU patient outcomes (improvement, exitus, etc.); and second, implementing these models within a Federated Learning framework to address the challenge of data sharing among different hospitals.
Objectives
The objectives are focused on: Computational models to be applied in risk estimation and decision support systems in ICUs. FL framework that allows the development of a joint model without the need to exchange data.
Use case
The starting point was the RegCovid19 project, where UBIKARE developed a platform enabling over 80 hospitals across Spain to share data about COVID patients admitted to ICUs. These data are processed and used to feed the predictive models, in which two typologies are considered: models with high explanatory capacity (vital in a healthcare environment) and models with lower explainability but higher predictive capacity, which are those used for the development of an FL framework.
Infrastructure
Hibrida On Premise Cloud
Technology
Machine learning and deep learning
Data
A private and anonymised dataset from the RegCovid19 project, which includes information on approximately 2500 patients during their entire stay in the ICU.
Resources
Researchers specialised in the development of AI models and Federated Learning frameworks. NAIHA platform from which the company accesses the stored data.
Difficulties and learning
The main problems encountered were: Data structuring challenges due to data heterogeneity. Necessity for a data sharing protocol due to data sensitivity. Necessity for balancing model explainability with predictive capacity.
KPIs (business impact and metrics of the model)
The main impacts on the business include: Positive outcomes for patients and healthcare professionals. Cost reduction and enhanced healthcare system efficiency. Support for advancements in digital health innovations.
Funding
HAZITEK Estratégico AIXIA (2022-2024)
Collaborators, Partners
Ikerlan S. Coop. UBIKARE ZAINKETAK, S.L
