@inproceedings{gangavarapu-2026-hgkai26,
    title = "hgkai26 at {MEDIQA}-{EVAL} 2026: Automated Evaluation of Visual Medical Question Answering Using {LLM}-as-a-Judge",
    author = "Gangavarapu, Haritha",
    editor = "Ben Abacha, Asma  and
      Bethard, Steven  and
      Bitterman, Danielle  and
      Naumann, Tristan  and
      Roberts, Kirk",
    booktitle = "Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical {NLP}) @ {LREC} 2026",
    month = may,
    year = "2026",
    address = "Palma, Mallorca (Spain)",
    publisher = "ELRA Language Resources Association (ELRA)",
    url = "https://preview.aclanthology.org/paragraph-normalization/2026.clinicalnlp-1.29/",
    doi = "10.63317/4n9skmf9rive",
    pages = "257--261",
    abstract = "As there is a rise in the use of multimodal large language models (LLMs) for medical response generation, it is necessary to have reliable automated evaluation mechanisms that can assess the quality of model-generated outputs. The MediQA-Eval 2026 shared task focuses on grading AI-generated dermatology and wound care responses using structured human-aligned rubrics. In this work, we explore a zero-shot multimodal LLM-as-a-Judge framework to assess candidate responses across multiple quality dimensions. System performance is evaluated using the official task metrics designed to reflect alignment with human judgments. Our findings provide preliminary insights into the feasibility and limitations of LLM-based evaluators for rubric-guided medical response assessment."
}