Artificial Intelligence in Rare Disease Phenotyping: A Systematic Review Protocol

Background: Rare diseases (RDs) diagnosis remains a significant challenge due to their heterogeneity, lack of data and limited clinical awareness. The identification of observable signs and symptoms in RDs – known as phenotyping - is a critical step for the diagnostic process. Advanced Artificial Intelligence (AI) is a promising tool for addressing efficient RDs phenotyping, potentially assisting researchers, physicians and experts into RD management. However, the broad landscape of AI applications in this area is fragmented and the current development state is still unclear. Objectives: This paper describes a systematic review protocol aimed at mapping, examining and summarising the current state of AI-based applications for RD phenotyping. Methods: This protocol was defined according to the PRISMA 2020 statement. Literature search will be performed in a comprehensive and reproducible way across 3 well-stablished databases: Medline (Pubmed), Web of Science, Scopus, ACM and IEE Xplore. Key steps are further defined in this protocol, including research questions, eligibility criteria, study selection process, data extraction and quality appraisal, as well as quantitative and qualitative data analysis methods. Expected Results: This systematic review is expected to provide up-to-date state-of-the-art and an overview of effective data sources and AI-based applications for RD diagnosis through phenotyping, further describing strengths and limitations of current approaches. Conclusions: We expect that this review will potentially guide future initiatives aimed at developing practical and effective AI-based tools for RD phenotyping, which corresponds to the process commonly used by physicians and experts in their daily practice.