Search

Information Extraction (IE) is a core task in Natural Language Processing (NLP) where the objective is to identify factual knowledge in textual documents (often unstructured), and feed downstream use cases with the resulting output. In genomic medicine for instance, being able to extract the most precise list of phenotypes associated to a patient allows to improve genetic disease diagnostic, which represents a vital step in the modern deep phenotyping approach. As most of the phenotypic information lies in clinical reports, the challenge is to build an IE pipeline to automatically recognize phenotype concepts from free-text notes. A new machine learning paradigm around large language models (LLM) has given rise of an increasing number of academic works on this topic lately, where sophisticated combinations of different technics have been employed to improve the phenotypes extraction accuracy. Even more recently released, the ChatGPT 1 application nevertheless raises the question of the relevance of these approches compared to this new generic one based on an instruction-oriented LLM. In this paper, we propose a rigorous evaluation of ChatGPT and the current state-of-the-art solutions on this specific task, and discuss the possible impacts and the technical evolutions to consider in the medical domain.Clinical relevance- Deep phenotyping on electronic health records has proven its ability to improve genetic diagnosis by clinical exomes [10]. Thus, comparing state-of-the-art solutions in order to derive insights and improving research paths is essential.

Introduction: Multiple sclerosis (MS), a neuroinflammatory and demyelinating disease, modifies the normal connectivity among different brain regions involved in specific functions. Functional magnetic resonance imaging (fMRI), based on local changes in oxygen level as a response to the increase in neural activity, provides an approach to neural connectivity and brain dynamics which give us an overview on visual, motor and cognitive dysfunction and their mechanisms., Development: An advanced search was performed using PubMed. Terms 'fMRI', 'visual', 'motor', 'cognitive' and 'multiple sclerosis' included in title and abstract were considered. We focus on original articles available in English. Articles were included based on their abstracts, looking for those potentially useful for understanding functional changes in MS. An important amount of studies have used fMRI as a complementary tool in the study of MS and clinically relevant alterations compromising visual, motor and cognitive domains. Since the earliest stages of the disease, local activity, and global neural dynamics appear to be compromised. Even when functional performance is still preserved, a different recruitment of neural resources arises as a compensatory response to disconnection observed in the disease., Conclusions: The main findings of fMRI applied to MS are strongly related to the demyelinating nature of the disease and provide an adequate insight into the mechanisms that underlie functional alterations reported in this disease. fMRI also appears to be useful for studying disease evolution and response to treatment in MS and other disorders.