Multimodal and federated learning

Project objectives

The aim of the project is to develop an Artificial Intelligence (AI) system based on Multimodal Learning (ML), using a Federated Learning (FL) paradigm and that is transparent. Such an AI system will be the necessary bridge between current clinical practice and personalized medicine, and will be applied to predict the clinical outcome of non-communicable diseases, such as non-small cell lung cancer. The project will allow to identify quantitative biomarkers calculated from heterogeneous data (e.g. images, EHR data, etc.) to represent the phenotype of the disease, providing a prognostic signature, opening the possibility of implementing personalized treatment.

The above objective addresses the three major challenges that AI faces today in the healthcare sector. The first is a consequence of the fact that data often exists in isolation. Traditional AI machine learning approaches need to combine all data in one place, but recent scientific evidence shows that FL can become the foundation of next-generation machine learning, being GDPR compliant.

The second challenge is motivated by the poor prognosis associated with many diseases. This creates a sense of urgency for the use of AI in this field, which has fostered the research of several omics sciences, such as genomics and radiomics, which however have shown limitations when investigated alone. Indeed, since the interpretation of clinical data is multimodal by its very nature, it is necessary to develop ML-based decision support systems to obtain a richer representation of the information available in different modalities, progressing towards a more informative and performant clinical decision-making process. At the same time, it is necessary to define methods to explain algorithmic decisions to improve trust and transparency in AI methods.

The third challenge concerns the identification of new quantitative biomarkers that can integrate heterogeneous information to describe the phenotype of the disease pursuing the paradigm of 5P medicine.

By addressing these challenges, the project promotes sustainability, responsible application of AI, interoperability and sharing of patient data within the AI ​​ecosystem in the health sector, to improve the quality of life and quality of care provided to patients, making them smarter and more intelligible.

Start and end date

Project Manager

Ing. Ermanno Cordelli, RTD based on the project

Coordinating institution of the project

Other Institutions involved

• Italian Diagnostic Center SpA

Funding source(s).