Applications in Clinical Data Science with focus on Multimodal Deep Learning methods

Clinical Data Science (CDS) represents the intersection between Health Informatics and the usage and design of advanced Data Analytics tools to solve critical healthcare services problems. The healthcare technology market trends reflect the urgent need for advanced digital solutions. The global healthcare informatics market was valued at USD 39.45 billion in 2016 and is expected to reach USD 123.24 billion by 2025. One of the critical points to develop solutions concerns the usage of high-quality input data. In the context of CDS, 80% of this data is locked in unstructured medical notes. Therefore, Natural Language Processing based tools streamline clinical research workflows and will help increase the value of patient data. In this Ph.D., we will research theoretical knowledge and application-oriented solutions in the domain of Clinical Data Science with a specific focus on Deep Learning methods. Specifically, we want first to develop an open-source tool that transforms standard Electronic Health Records (EHRs) into embeddings for popular ML packages. Then extend, modify, or even define new embeddings to model different types of EHRs to later merge the unstructured medical notes with structured data to define a Generalized Healthcare Transformer architecture, extending previous work such as ClinicalBERT (Huang et al., 2019).