Grant: Project Grant

In an increasingly data-driven world, artificial intelligence is expected to be a key tool for converting the big data into tangible benefits. Machine learning, a sub-field of artificial intelligence, aims to identify complex patterns in multi-dimensional data and use these uncovered patterns to classify or make predictions on new unseen cases. However, many datasets are typically needed for training of the advanced machine learning models to produce highly accurate results. This is also the main limiting factor why these advanced machine learning models cannot unfold their full potential in health care applications if personal or other sensitive data is used that data owners cannot or will not share for training of a machine learning model at a central location. Pediatric brain tumors are the most common solid tumor in children. Among others, treatment and prognosis depend on the tumor type, the location within the brain, whether it has spread, as well as clinical parameters. Due to the numerous different types of pediatric brain tumors and treatment options, it is especially challenging to collect enough datasets needed to train robust machine learning models that produce accurate results for children with brain tumors. The aim of this proposed research is to develop a novel distributed learning system for training of advanced machine learning models for pediatric brain tumors to overcome the limiting factors of data collection and analysis in a centralized fashion. The main goal of the distributed learning architecture is that the individual patient data does not leave the contributing clinical centers but are only used locally to train the machine learning models, which do not contain any patient data at the individual level anymore. In this research project, we will invent, develop, implement, and evaluate multiple approaches for distributed machine learning model training that have the potential to improve diagnosis and treatment of children with brain tumors.