louis-dataset
Description
The LouiS Validation Dataset was created to support research, education, and evaluation of systems for stroke syndrome recognition and neuroanatomical localization. It consists of two components: a structured database of stroke syndromes and a small set of curated clinical cases. The dataset represents expert reference knowledge derived from neurology textbooks and peer reviewed articles and does not contain patient level data. The Stroke Syndromes Database includes 619 manually curated entries. Each entry defines a stroke syndrome using standardized fields for clinical features, anatomical localization, vascular territory, and symptom lateralization. Lateralization is encoded using fixed categories to allow consistent interpretation. Syndrome names are provided in English, Portuguese, and Spanish, with translations generated using language models and manually verified. Descriptions were reconstructed by consolidating multiple sources to improve consistency. The dataset also includes 12 curated clinical cases with reference diagnoses, used for validation and benchmarking. Data are provided in JSON and CSV formats. The dataset is intended as a transparent knowledge base for retrieval based systems, education, and evaluation of diagnostic tools in stroke neurology.
Files
Steps to reproduce
All entries were reviewed for clinical coherence and internal consistency, then compiled into a structured dataset using custom validation and formatting scripts. The dataset was exported in JSON and CSV formats and used as the knowledge base for the LouiS retrieval augmented generation workflow. This process is reproducible using the same literature sources, curation schema, manual reconstruction steps, and translation procedure.
Institutions
- Universidade de Sao Paulo Hospital das Clinicas da Faculdade de Medicina de Ribeirao Preto