Topology-Aware Query Selection for Surgical Instrument Instance Segmentation

2026-08-11

AI Summary

This work addresses the inconsistency in surgical instrument instance segmentation caused by duplicated, fragmented, or missed detections by formulating instance selection as a relational variable-cardinality set prediction task for the first time. Building upon fixed candidate masks generated by Mask2Former, the method constructs a complete graph to explicitly model node and edge relationships, integrating graph neural networks, geometric relation encoding, and combinatorial optimization to solve for a topology-aware structured subset. Evaluated on a test set of 22 cases, the approach improves instance-level F1 scores by 0.0504–0.0612 and reduces positive-frame failure rates by 0.0848–0.1060. Consistent performance gains are further validated across three subsets within the ROBUST-MIPS domain, significantly enhancing both the count accuracy and identity consistency of segmented instances.

Abstract

Accurate foreground masks can still form an incorrect surgical-instrument instance set: duplicate, fragmented, merged, missed, or empty-frame predictions may preserve favorable pixel overlap while violating object identity and count. Final query selection is therefore a relational, variable-cardinality problem rather than a collection of independent candidate decisions. We evaluate topology-aware query selection, which represents the nonempty candidates of a fixed Mask2Former as a complete graph, learns relational candidate and pair representations, predicts set cardinality, and solves an exact structured subset problem. The formal comparison is the complete relational path versus a node-feature-matched path; it evaluates the combined effect of pairwise geometry, message passing, and the additional relational-path capacity, not an isolated component. On the sealed 22-case source test, all three discovery seeds supported instance-set performance improvement with segmentation fidelity and predefined technical-safety preservation: instance F1 increased by 0.0504--0.0612 and positive-frame set-failure rate decreased by 0.0848--0.1060. Direct ROBUST-MIPS transfer reproduced the complete result in all three seeds. Endoscapes supported only one of three seeds and therefore did not establish stable direct transfer. Taken together, the results support a bounded conclusion: the evaluated complete path improved coherent instance-set construction from fixed Mask2Former candidates in specified native-instance contracts, while stable cross-domain transfer and component-specific effects remain unestablished.

Problem

• instance segmentation
• surgical instrument
• query selection
• object identity
• set cardinality

Innovation

• topology-aware
• instance segmentation
• relational reasoning
• query selection
• structured subset optimization