SUREON: A Benchmark and Vision-Language Model for Surgical Reasoning

AI Digest

SUREON 是首个针对手术视觉推理的基准测试与模型，揭示现有模型和专家表现均不足。

SUREON introduces a benchmark and VLM for surgical reasoning, showing current models and experts struggle with visual inference.

Key points

• SUREON 通过对抗性过滤和多模型精炼构建视觉-语言数据集 SUREON constructs a visual-language dataset via adversarial filtering and multi-model refinement
• 现有模型和专家外科医生在手术推理任务中准确率均低于65% Current models and surgeons achieve <65% accuracy on surgical reasoning tasks
• SUREON-VLM 在多个基准测试中超越开源模型并匹配专科模型 SUREON-VLM outperforms open-source models and matches specialist models
• 训练数据精简后视觉接地能力提升29.1个百分点 Visual grounding improves 29.1% with curated training data

Takeaway: 手术视觉推理仍属难题，SUREON 提供了新的评估框架和改进方向。 / Surgical visual reasoning remains challenging, with SUREON offering new evaluation and improvement frameworks.

Why it matters 提供手术AI评估新范式，揭示视觉推理技术瓶颈与突破路径。