Advancing Complex Medical Communication in Arabic with Sporo AraSum: Surpassing Existing Large Language Models

  • 2024-11-20 18:10:19
  • Chanseo Lee, Sonu Kumar, Kimon A. Vogt, Sam Meraj, Antonia Vogt
  • 0

Abstract

The increasing demand for multilingual capabilities in healthcare underscoresthe need for AI models adept at processing diverse languages, particularly inclinical documentation and decision-making. Arabic, with its complexmorphology, syntax, and diglossia, poses unique challenges for natural languageprocessing (NLP) in medical contexts. This case study evaluates Sporo AraSum, alanguage model tailored for Arabic clinical documentation, against JAIS, theleading Arabic NLP model. Using synthetic datasets and modified PDQI-9 metricsmodified ourselves for the purposes of assessing model performances in adifferent language. The study assessed the models' performance in summarizingpatient-physician interactions, focusing on accuracy, comprehensiveness,clinical utility, and linguistic-cultural competence. Results indicate that Sporo AraSum significantly outperforms JAIS inAI-centric quantitative metrics and all qualitative attributes measured in ourmodified version of the PDQI-9. AraSum's architecture enables precise andculturally sensitive documentation, addressing the linguistic nuances of Arabicwhile mitigating risks of AI hallucinations. These findings suggest that SporoAraSum is better suited to meet the demands of Arabic-speaking healthcareenvironments, offering a transformative solution for multilingual clinicalworkflows. Future research should incorporate real-world data to furthervalidate these findings and explore broader integration into healthcaresystems.

 

Quick Read (beta)

loading the full paper ...