Global and local relational reasoning enable scene understanding models to perform human-like scene analysis and understanding. Scene understanding enables better semantic segmentation and object-to-object interaction detection. In the medical domain, a robust surgical scene understanding model allows the automation of surgical skill evaluation, real-time monitoring of surgeon’s performance and post-surgical analysis. This letter introduces a globally-reasoned multi-task surgical scene understanding model capable of performing instrument segmentation and tool-tissue interaction detection. Here, we incorporate global relational reasoning in the latent interaction space and introduce multi-scale local (neighborhood) reasoning in the coordinate space to improve segmentation. Utilizing the multi-task model setup, the performance of the visual-semantic graph attention network in interaction detection is further enhanced through global reasoning. The global interaction space features from the segmentation module are introduced into the graph network, allowing it to detect interactions based on both node-to-node and global interaction reasoning. Our model reduces the computation cost compared to running two independent single-task models by sharing common modules, which is indispensable for practical applications. Using a sequential optimization technique, the proposed multi-task model outperforms other state-of-the-art single-task models on the MICCAI endoscopic vision challenge 2018 dataset. Additionally, we also observe the performance of the multi-task model when trained using the knowledge distillation technique. The official code implementation is made available in GitHub.
Several approaches have been introduced to understand surgical scenes through downstream tasks like captioning and surgical scene graph generation. However, most of them heavily rely on an independent object detector and region-based feature extractor. Encompassing computationally expensive detection and feature extraction models, these multi-stage methods suffer from slow inference speed, making them less suitable for real-time surgical applications. The performance of the downstream tasks also degrades from inheriting errors of the earlier modules of the pipeline. This work develops a detector-free gradient-based localized feature extraction approach that enables end-to-end model training for downstream surgical tasks such as report generation and tool-tissue interaction graph prediction. We eliminate the need for object detection or region proposal and feature extraction networks by extracting the features of interest from the discriminative regions in the feature map of the classification models. Here, the discriminative regions are localized using gradient-based localization techniques (e.g. Grad-CAM). We show that our proposed approaches enable the real-time deployment of end-to-end models for surgical downstream tasks. We extensively validate our approach on two surgical tasks: captioning and scene graph generation. The results prove that our gradient-based localized feature extraction methods effectively substitute the detector and feature extractor networks, allowing end-to-end model development with faster inference speed, essential for real-time surgical scene understanding tasks. The code is publicly available at https://github.com/PangWinnie0219/GradCAMDownstreamTask .