Objective-Aligned Direct Answer SFT for Robust Multi-Frame Medical VQA

We describe our submission to the MedReason 2026 challenge, covering multiple-choice (MCQ) and open-ended (OE) medical visual question answering (VQA) under fully offline, containerized inference. Our first finding is that MCQ retrieval must compare answer \emph{semantics} rather than answer labels: labels are independently assigned per question, so copying a retrieved neighbor's label transfers no useful information, whereas scoring each current option's text against correct-answer text from similar training cases raises retrieval-only accuracy from 20.0% to 57.5% on a 200-case retrieval-excluded development holdout. Our second finding attributes the submitted system's accuracy: holding the task-specific MCQ Low-Rank Adaptation (LoRA) adapter fixed and varying the number $k$ of in-prompt retrieved examples changes accuracy by at most one case --- 187/200 (93.5%) at both $k=0$ and the adapter's training-time $k=1$, 188/200 (94.0%) at the packaged runtime's default $k=3$ --- and the submitted confidence-gated override adds no net accuracy on top of $k=3$, selecting the VLM in 198/200 cases. With the final MCQ adapter fixed, retrieval changes accuracy by at most one case, and gating provides no net gain. On 20 OE cases, token-F1 and RaTEScore~\cite{zhao2024ratescore} decrease as $k$ grows, but paired sign tests on token-F1 differences are nonsignificant ($p \ge 0.29$); a single-annotator comparison found 6/20 wrong-anchor errors for the final configuration and 14/20 for an earlier configuration that jointly differed in routing, adapter, and prompting. The system reaches 94.0% MCQ accuracy on the development holdout and 93.20% on the organizer's official pre-evaluation, versus 29.43% for the off-the-shelf reference baseline, while both of the organizer's open-ended scores are lower than that baseline's (ground-truth agreement 1.245 versus 1.588, visual accuracy 1.995 versus 2.696, each out of 4).

Multimodal large language models (MLLMs) applied to Medical Visual Question Answering (VQA) tend to produce overconfident outputs regardless of actual correctness, and existing verbalized confidence calibration methods, developed primarily for text only LLMs, do not account for the multimodal nature of medical image understanding. This work proposes a training based framework that finetunes MLLMs to improve their calibration using a composite loss function combining a Brier style calibration term, an anchor regularizer that prevents confidence collapse toward extreme values, a contrastive image text alignment term, and a KL based model stabilization term. The alignment signal is derived from a $2 \times 2$ factorial perturbation design that crosses image presence with text integrity, probing the reliance of the model on visual modality input versus language priors. Finally, a top K KL divergence regularizer is used to protect the answering ability of the model during finetuning. Across three Medical VQA benchmarks and two architectures (MedGemma 4B IT and Qwen2 VL 7B Instruct), our method reduces calibration error by 60% or more, and improves discrimination by 26% or more, while preserving predictive accuracy. On average across benchmarks, the technique outperforms prompting based, sampling based, and training based approaches, and ablation experiments confirm that each component of the loss function is indeed necessary for improving the calibration. All code for the experiments is publicly available.

Multi-image medical VQA is not merely a prompt-length problem; it is a fundamental challenge of agentic decision-making. Medical vision-language agents must aggregate evidence across ordered images, remain robust to answer-order perturbations, and avoid overfitting to noisy search-time feedback. We study MedFrameQA through a controlled comparison of five inference-time agentic strategies, optimized using the same high-budget ShinkaEvolve configuration and evaluated on a reproducible internal frozen split (1,331 evolution, 665 holdout, 855 final test). Across five independent repeated runs, the strongest method emerges as the simplest robust aggregator: the \textbf{order-vote} policy achieves $57.89 \pm 0.65\%$ final-test accuracy, significantly outperforming the fixed baseline ($52.73 \pm 0.42\%$) and the more complex, albeit brittle, order-rerank variant ($55.79 \pm 0.43\%$). Paired bootstrap analysis confirms these significant gains. Extending the evolutionary search budget from 50 to 100 generations yields no generalization benefit: while holdout performance marginally increases, final-test accuracy drops from $57.89\%$ to $56.02\%$. Our findings suggest that for multi-image medical reasoning, defining the correct agentic decision rule is substantially more impactful than expanding the optimization search budget.