Artificial Intelligence for Precision Medicine in COVID-19

Project objectives

The COVID-19 pandemic has caused millions of cases and deaths, and the artificial intelligence (AI) community, after focusing on pathology detection in medical images, is now focusing its efforts on developing methods for analyzing risk factors to predict disease progression. Advances in AI, including deep learning (DL), are expected to bring significant contributions in this area: however, most DL models only consider clinical information or data extracted from images, and few works have studied the fusion between quantitative biomarkers extracted from radiological images and multivariate clinical data. To this end, this project aims to advance multimodal DL (MDL), an area of ​​great interest in its infancy, by addressing currently open questions: shared data representations, what is the optimal fusion architecture, regularization, learning in the case of missing data and modalities. Furthermore, the project explores methods to explain MDL decisions, generating textual and visual explanations and concepts understandable even to a non-expert, a topic not yet addressed in the literature. These explanations will help clinicians, patients, and regulators trust the results of AI models, including understanding what risk factors are and why they provide useful information to predict disease progression. Using publicly available datasets, including chest x-rays (CXR), chest computed tomography (CCT), and clinical information, our AI-based risk analysis methodology will result in a clear and concise set of quantitative biomarkers to identify patients at risk for severe outcomes, so the right therapy can be selected for the right patient.

Start and end date

Project Manager

Coordinating institution of the project

Other Institutions involved

• Shenzhen University, School of Computer Science and Software Engineering

Funding source(s).