Published Paper Details:

Stroke is one of the major causes of death and permanent disability in the world, and it generates an astronomical amount of scientific information that is ever expanding. The current practice of manually synthensizing this information is proving more difficult and it is a barrier to achieving timely insight generation and evidence-based decision making. This research presents a new machine reading framework that combines LDA topic modeling with Open IE to use stroke-related biomedical literature published between 2000 and May 2025 in an arteriosclerosis-related randomized controlled trial. This research curated a corpus of 179,219 PubMed abstracts by using stroke-specific MeSH terms and using the biomedical natural language processing tools to preprocess text and recognize an entity. Ten significant thematic clusters that consist of neuroimaging, acute reperfusion therapies, neuroinflammation, rehabilitation, and AI-driven stroke prediction became apparent in the analysis. Open IE captured over 4.1 million subject-predicate-object triplets, allowing querying by structure around comorbidities, the performance of drugs, and currently emerging risk variables like COVID-19. The current study noted an extreme increase in the number of studies that apply deep learning and large language models (LLMs) to predict stroke after 2022. Topic coherence scores and expert reviews were used in validating the model to ascertain the reliability of results. This was lastly achieved through the creation of an interactive web-based dashboard allowing them to visualize topic trends and explore the extracted knowledge to provide researchers with a dynamic way to discover literature. This paper proves the capability of machine reading to make literature review scalable and data driven and thus assist in faster discovery and better clinical decision-making in stroke research.