arXiv:2610.09800v1 Announce Type: new
Abstract: Surgical video generation holds substantial potential for surgical education, simulation, and data augmentation, yet generating surgical videos with re...
By Tsz-Yui Qin, Siyu Zhou, Chi-Keung Tang, Yuxiang Nie, Shu Yang
InnerLogic‑Surg is an open family of transformer‑based surgical vision encoders, ranging from 21 M to 300 M parameters, that are continually pretrained on millions of surgical images drawn from 11,965 internet‑sourced videos covering roughly 2,900 hours and more than fifteen procedure types. The training pipeline builds on DINOv3 to maintain general representation quality while adding a learned feature‑upsampling module for efficient high‑resolution dense prediction. On a benchmark of 20 surgical tasks—including workflow, instrument and anatomy recognition, segmentation, detection, and monocular depth—the largest InnerLogic‑Surg model achieves the best linear‑probe performance across all tasks, demonstrating that surgical continual pretraining yields transferable representations that generalize across procedures, anatomical regions, and task types.
InnerLogic‑Surg is an open family of transformer‑based surgical vision encoders ranging from 21 M to 300 M parameters, trained by continually pretraining on millions of surgical images from 11,965 internet‑sourced videos covering ~2,900 hours and more than fifteen procedure types. The training pipeline builds on DINOv3 to preserve general representation ability while adding a learned feature‑upsampling module for high‑resolution dense prediction at low token cost. In a benchmark covering workflow, instrument and anatomy recognition, segmentation, detection, and monocular depth, the largest InnerLogic‑Surg model achieves the best linear‑probe performance across all 20 surgical tasks, demonstrating transferable representations that generalize across procedures, anatomical regions, and task types.
By Kun Yuan, Hao Ding, Yiqing Shen, Gasper Podobnik, Benjamin Killeen, Tito Porras, Axel Krieger, Mathias Unberath
SurgAtlas is the largest surgical video‑language dataset, containing 15,291 videos (2,391 hours) across 18 specialties and over 5,000 procedure types, all sourced from public YouTube. It uniquely includes open‑surgery videos at scale (6,182) alongside more than 9,000 minimally invasive recordings, and introduces standardized benchmarks for open‑surgery video understanding. The dataset offers a rich, multi‑tier annotation schema—segment‑level captions, step/phase descriptions, video‑level surgical narratives, and reasoning‑oriented VQA pairs—validated by experts and built through an automated LLM‑enriched pipeline.
"whyItMatters":"SurgAtlas provides an unprecedentedly large, diverse, and clinically validated resource that can train and benchmark multimodal surgical AI models, advancing the development of next‑generation foundation models for surgery."
By Filippos Bellos, Andre S. Gala-Garza, Miaowei Wang, Alyssa M. Hardin, Ahmad M. Hider, Li Yayuan, Jing Bi, Susan Liang, Chenliang Xu, Donald S. Likosky, Jason J. Corso
arXiv:2603.29962v4 Announce Type: replace
Abstract: Surgical procedures are inherently complex and risky, requiring extensive expertise and constant focus to navigate evolving intraoperative scenes....
By Shi Li, Vinkle Srivastav, Nicolas Chanel, Saurav Sharma, Nabani Banik, Lorenzo Arboit, Kun Yuan, Pietro Mascagni, Nicolas Padoy
We introduce SurgAtlas, the largest surgical video-language dataset to date, comprising 15,291 videos (2,391 hours) spanning 18 surgical specialties and over 5,000 procedure types, sourced entirely from publicly available YouTube content. SurgAtlas is also the first surgical video-language dataset to include open surgery at scale, with 6,182 open procedure videos alongside over 9,000 minimally invasive recordings, and the first to establish standardized benchmarks for open-surgery video understanding.
The paper presents a method for localizing functional surgical landmarks—specifically instrument tips and anchors—in surgical videos without requiring manual pixel-level mask annotations. It leverages vision foundation models, such as SAM 3, to generate dense structural priors through zero‑shot, point‑prompted masks, and refines landmark predictions with a lightweight, coarse‑to‑fine multi‑frame network. Experiments on 7,867 clips from 60 videos show that the approach achieves F1 scores of 72.4% for tip and 58.0% for anchor localization, with ablations confirming the benefits of structural priors and refinement stages.
By Chenyan Jing, Hao Ding, Lalithkumar Seenivasan, Jacob M. Delgado L\'opez, Mathias Unberath
Effective multi-task learning for surgical scene understanding is fundamentally hindered by annotation granularity mismatch; temporal workflow tasks such as phase recognition, step recognition and anticipation benefit from dense frame-level supervision, whereas pixel-level spatial tasks including instrument segmentation and action recognition are only sparsely annotated on selected keyframes due to prohibitive labeling costs. This supervision imbalance undermines shared representation learning and limits joint optimization across heterogeneous surgical tasks.
arXiv:2607.19889v3 Announce Type: replace
Abstract: Recognizing instrument-tissue interactions is essential for context-aware surgical AI. Vision-language models offer a natural way to inject semanti...
By Jiajun Cheng, Sainan Liu, Subarna Tripathi, Xiaofan Yu, Shan Lin
arXiv:2608.30872v1 Announce Type: new
Abstract: Objective assessment of surgical technical skill is important for surgical training and structured feedback, but current workflows remain dependent on...
By Chaohui Dang, Zheheng Jiang, James Glasbey, David Luke, Theodoros Arvanitis, Le Zhang
The paper introduces Latent-Action-Guided Video-Language Feature Learning (LAG-VLFL) for recognizing surgical instrument–tissue interactions. By compressing frame-to-frame feature changes into latent actions and predicting next‑frame features, the method aligns video and textual action descriptions without requiring extra spatial or motion annotations. Experiments show that LAG-VLFL improves interaction grounding, temporal‑direction sensitivity, and achieves competitive recognition with faster inference and lower INT4 accuracy loss compared to V‑JEPA2/2.1.
By Jiajun Cheng, Sainan Liu, Subarna Tripathi, Xiaofan Yu, Shan Lin
The paper introduces SurgFUTR, a state‑change learning framework for predicting future events in endoscopic videos. Instead of forecasting raw observations, it classifies transitions between current and future states using a teacher‑student architecture and an Action Dynamics module. The authors also present SFPBench, a benchmark with five short‑ and long‑term prediction tasks, and demonstrate consistent improvements across multiple datasets and procedures, including cross‑procedure transfer.
By Saurav Sharma, Chinedu Innocent Nwoye, Didier Mutter, Nicolas Padoy