SurgPace builds smaller smarter AI for robotic surgery tasks
SurgPace: Complexity-Aware Progressive Distillation for Surgical Multimodal Large Language Models
Computer Vision and Pattern Recognition
Summary
AI models that understand images and text together can help during robotic surgeries, but they are usually too big and slow for real use. The authors developed SurgPace, a way to teach smaller AI models by starting with easy surgical examples and gradually moving to harder ones, making learning smoother and more stable. Their method also changes how the AI is taught over time, focusing first on copying a big teacher AI and later on directly learning the task. SurgPace works better and faster than other methods on large surgical datasets.
What this means in practice
- •For robotic surgery system developers: Create more efficient AI assistants that understand surgical scenes and questions without heavy computing demand during operations.
- •For medical imaging software engineers: Integrate smaller multimodal AI models trained with difficulty-based curricula to improve accuracy and speed in surgical image analysis tools.
Authors
Zhicheng He, Xiaohan Xing, Hongqiu Wang, Kun Yuan, Guankun Wang, Yanheng Li, Haoxuan Che, Nassir Navab, Hongliang Ren, Long Bai
Abstract
Multimodal large language models (MLLMs) have shown strong potential in robotic surgery for question answering, scene understanding, and grounded assistance, but their high computational cost limits real clinical deployment. Knowledge distillation provides a practical way to build compact surgical MLLMs, yet direct distillation is not well suited to surgical data because surgical samples vary greatly in difficulty, visual ambiguity, and required clinical context, making uniform teacher-student transfer unstable. In addition, the role of supervision changes during training: early stages require stronger teacher guidance for multimodal alignment, while later stages rely more on task supervision for effective optimization. To address these challenges, we propose SurgPace, a complexity-aware continuous curriculum framework for progressive distillation of surgical vision-language models. SurgPace estimates sample complexity using predictive confidence and Lorentz geometric consistency, and uses the fused score to guide a continuous easy-to-hard training process. We further introduce hierarchical representation alignment, which aligns the student with the teacher at the feature, logit, and probability levels, while gradually shifting the training focus from teacher imitation to task supervision. This design separates sample-level curriculum from objective-level scheduling, leading to more stable and effective distillation. We evaluate SurgPace on Surg-396K, a benchmark built from four public surgical datasets and covering five dialogue paradigms. SurgPace consistently outperforms strong MLLM distillation baselines, demonstrating its effectiveness for building compact, memory-efficient surgical MLLMs with faster decoding under controlled benchmarks. Code is available at https://github.com/hezhicheng2002/SurgPace.