Search papers, labs, and topics across Lattice.
This paper introduces TCSR-Monitor, a novel post-hoc failure-monitoring framework designed to enhance surgical segmentation networks by integrating confidence with shape, temporal consistency, and image quality cues. The method addresses the critical issue of silent failures that occur under acquisition degradation, where models may exhibit high confidence despite incorrect predictions. Evaluation on the EndoVis 2017 dataset demonstrates that TCSR-Monitor significantly outperforms traditional confidence-based monitoring approaches, although it still faces challenges with false alarms and transferability to new models.
TCSR-Monitor reveals that relying solely on confidence can lead to silent failures in surgical segmentation, highlighting the need for a multi-faceted monitoring approach.
Surgical segmentation networks can fail silently under acquisition degradation: predicted masks may be wrong even when model confidence remains high. Existing deployment-time monitors rely primarily on uncertainty estimates and can therefore miss confident failures. We present TCSR-Monitor (Temporal Conformal Surgical Risk Monitor), a post-hoc failure-monitoring framework that combines confidence with observable shape, temporal-consistency, and image-quality cues. TCSR-Monitor wraps a frozen segmentation model, requires no model internals, and operates without ground truth at deployment. We also introduce a validation protocol to assess whether alarms remain credible under distribution shift. On EndoVis 2017, leave-one-corruption-out evaluation shows that TCSR-Monitor generalizes to unseen acquisition degradations and substantially outperforms confidence-based baselines. A circularity control confirms that it predicts segmentation failure rather than simply detecting corrupted images. Mondrian conformal calibration balances miss-rates across degradation severities, but a single global threshold still produces false alarms on up to 40% of correctly segmented frames at moderate corruption. Zero-shot transfer to SAM2 demonstrates feature portability, although entropy outperforms the transferred monitor at both evaluated thresholds. Overall, reliable monitoring under acquisition degradation benefits from complementary observable signals beyond confidence alone, but substantial false-alarm and transfer limitations remain.
