Abstract

This study proposes a reasoning-based MLLM that visualizes the inference process to address the challenges of black-box behavior in Medical Visual Question Answering and to provide verifiable reasoning evidence for generated answers. Based on Qwen3.5-9B, we perform parameter efficient fine-tuning using 4-bit quantization and LoRA, and train the model to uniformly generate outputs for both multiple-choice questions and open-ended responses that include clinical reasoning. In evaluations on the MedReason 2026 benchmark, the model achieved a multiple-choice accuracy of 96.47% the open-ended GT of 1.791 and the open-ended VA of 2.393, demonstrating high reasoning accuracy and clinical validity.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedReason_003.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: https://openreview.net/forum?id=oVrA1mbO0w

BibTex

@InProceedings{KonSat_CLINICVQA_MICCAISAT2026,
        author = { Kondo, Satoshi AND Kasai, Satoshi},
        title = { { CLINIC-VQA: Reasoning-Aware MLLM with Explicit Clinical Reasoning Traces for Medical Visual Question Answering } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17262},
        month = {pending},
        page = {pending}
}


back to top