SurgLQA: Scalable Long-Horizon Surgical Video Question Answering

2026-05-18

AI Summary

Existing approaches to surgical video question answering struggle to model the dynamic evolution and causal dependencies inherent in long-duration procedures. To address this limitation, this work proposes a unified framework for long-form surgical video QA, introducing a novel Faithful Temporal Compression (FTC) mechanism to preserve fine-grained temporal details and a Temporally Anchored Multi-strategy Scaling (TMS) module for adaptive reasoning. The method constructs long-range representations grounded in intrinsic temporal cues and integrates a temporally anchored test-time inference strategy, substantially enhancing both efficiency and performance. Evaluated on the Colon-LQA and REAL-Colon-VQA benchmarks, the proposed approach demonstrates significant advantages in long-range reasoning tasks, confirming its effectiveness and scalability.

Abstract

Surgical Video Question Answering (VideoQA) provides a promising paradigm for dynamic intraoperative interpretation, enabling real-time decision support and context-aware retrieval in clinical environments. Nevertheless, existing approaches are predominantly restricted to images or short clips, limiting their ability to model long-range procedural dynamics and causal dependencies across extended surgical workflows. To address this challenge, we propose SurgLQA, a unified long-horizon VideoQA framework for scalable surgical reasoning. This framework incorporates Faithful Temporal Consolidation (FTC), which leverages intrinsic temporal cues to construct compact long-range representations while preserving fine-grained temporal fidelity. Further, we develop Temporally-Grounded Multi-Policy Scaling (TMS), an adaptive test-time inference paradigm that strategically adjusts policy-level reasoning capacity within temporally grounded contexts. To facilitate systematic evaluation, we restructured a long-duration colonoscopy VideoQA benchmark, Colon-LQA, and conducted extensive experiments on Colon-LQA and REAL-Colon-VQA. Experimental results demonstrate that our approach achieves consistent performance gains in long-range reasoning with temporally grounded inference. Code link: https://github.com/RascalGdd/SurgLQA.

Problem

• Surgical Video Question Answering
• long-horizon reasoning
• temporal dynamics
• causal dependencies
• scalable surgical reasoning

Innovation

• Long-horizon VideoQA
• Faithful Temporal Consolidation
• Temporally-Grounded Multi-Policy Scaling
• Surgical Video Understanding
• Scalable Reasoning