Surgical action triplet recognition plays a vital role in computer-assisted and anatomical surgery. A fine-grained formulation of this task aims to simultaneously identify the surgical instrument, operative verb, and anatomical target, which are represented as triplets in the form of . Existing methods face two key challenges: 1) Relying solely on the visual features of instruments fails to identify abstract functional operations. 2) Limited modeling of inter-entity relationships hampers accurate triplet recognition. To address these challenges, we propose DHR2-Net, a novel Dual-Hierarchical Relational Reasoning framework for surgical action triplet recognition. First, DHR2-Net explicitly models verbs as semantic interactions between instruments and targets to capture the underlying surgical intent. Then, we construct first-order (instrument-context and target-context) and high-order (instrument-verb-target) dependencies through dual-hierarchical relational reasoning to build clear internal associations of triplets. To further enhance feature representation during the high-order reasoning stage, we introduce an inter-sample high-order relational reasoning module (IHR2) that leverages relational cues shared across samples within each batch for more explicit relational modeling. Meanwhile, an auxiliary actor identity preservation module enables the final triplet-level embedding to capture both identity and relational semantics. We evaluate DHR2-Net on the CholecT45 dataset using 5-fold cross-validation and the CholecTriplet2021 challenge splits. The experimental results show that DHR2-Net achieves superior performance in fine-grained component recognition and interaction modeling, as well as robust accuracy in triplet recognition. The visualization analyses further highlight its strong clinical interpretability and potential for practical deployment in computer-assisted intervention systems.
He et al. (Thu,) studied this question.