2

This study proposes a novel fault diagnosis method for rotating machinery using spatial–temporal graph feature learning to overcome noise challenges. By transforming time–frequency heatmaps into time–frequency similarity graphs (TFSGs), the method captures structural information, processed by an improved GraphSAGE network (IGN) for classification. Validation on axial flow pump and gearbox datasets demonstrates improved fault detection accuracy under noisy conditions.

We propose a novel Residual Aligner-based Network (RAN) for deformable image registration, addressing challenges in capturing separate and sliding motions of organs. By introducing a Motion Separable backbone and a Residual Aligner module, RAN achieves state-of-the-art accuracy in unsupervised registration of abdominal and lung CT scans, with reduced model size and computational cost.

This study introduces a novel polarization-based method for removing surgical smoke to restore image visibility and improve tissue observation during surgeries. By analyzing the propagation of polarized light through a smoke-tissue medium, it identifies that cross-polarized channel data is less affected by smoke compared to co-polarized data. Using polarization difference calculations for different color channels, the method estimates transmission parameters to reconstruct clearer biomedical images.

This study proposes a robust augmented reality system for real-time tracking of a gamma probe to enhance intraoperative cancer resection and lymph node identification. A dual-pattern marker combining chessboard vertices and circular dots enables highly accurate pose estimation and tracking, validated with ground truth data from the OptiTrack system. The integration with augmented reality provides visual feedback to surgeons, improving localization of pathological sites and reducing mental workload during surgery. The method achieves precise tracking accuracy with minimal error, a large workspace, and better performance than previous hybrid markers.

We propose a self-adaptive motion-scaling mechanism for teleoperation in robotic surgery to improve precision and efficiency. The system dynamically adjusts the motion-scaling ratio based on three factors – situation awareness (real-time context), skill level awareness (operator proficiency), and task awareness (current task complexity). This allows precise control for intricate manipulations while enabling faster movements to distant targets, minimizing the need for clutching or repositioning.