Stabilizing Temporal Inference Dynamics for Online Surgical Phase Recognition

2026-05-11

AI Summary

This work addresses the temporal instability of online surgical phase recognition models, which, despite high frame-level accuracy, produce fragmented workflow interpretations due to error cascades triggered by early misclassifications and the absence of an evidence accumulation mechanism. To resolve this, the authors propose a unified framework: during training, a Temporal Error Cascade (TEC) loss mitigates error propagation, while at inference, an Evidence-Gated Transition Predictor (EGTP) enables state transitions based on accumulated evidence. Additionally, they introduce the Temporal Fragmentation Index (TFI) to quantify prediction stability. The method is backbone-agnostic and consistently enhances temporal coherence and reduces fragmentation on both Cholec80 and AutoLaparo datasets, while maintaining or slightly improving frame-level accuracy.

Abstract

Online Surgical Phase Recognition (SPR) models can reach high frame-wise accuracy, yet their predictions often lack temporal stability, fragmenting workflow understanding and reducing the reliability of downstream assistance. We show that this instability is not random noise but arises from two mechanisms: early misclassifications corrupt temporal feature states and propagate forward to form error cascades, and phase transitions follow evidence-accumulation dynamics whereas most online SPR systems rely on memoryless frame-wise decisions, making them sensitive to transient confidence fluctuations. We propose a unified Train-Inference-Evaluation framework that explicitly stabilizes temporal inference dynamics using model-agnostic, plug-and-play components. For training, the Temporal Error-Cascade (TEC) loss suppresses error onset and mitigates forward error propagation by stabilizing temporal feature evolution. For inference, the Evidence-Gated Transition Predictor (EGTP) enforces evidence-driven state transitions, allowing phase changes only when accumulated evidence exceeds a confidence boundary. For evaluation, we introduce the Temporal Fragmentation Index (TFI), a reliability-aware metric that quantifies instability-induced temporal disagreement beyond conventional frame-wise and token-based measures. Experiments on Cholec80 and AutoLaparo across three representative backbones show that the proposed framework substantially improves temporal stability and reduces prediction fragmentation, while maintaining or modestly improving frame-wise performance.

Problem

• Temporal Stability
• Surgical Phase Recognition
• Error Cascades
• Evidence Accumulation
• Prediction Fragmentation

Innovation

• Temporal Stability
• Error Cascade
• Evidence Accumulation
• Online Surgical Phase Recognition
• Temporal Fragmentation Index