25 March 2022 in IEEE Transactions on Neural Networks and Learning Systems
Percutaneous coronary intervention (PCI) has increasingly become the main treatment for coronary artery disease. The procedure requires high experienced skills and dexterous manipulations. However, there are few techniques to model PCI skill so far. In this study, a learning framework with local and ensemble learning is proposed to learn skill characteristics of different skill-level subjects from their PCI manipulations. Ten interventional cardiologists (four experts and six novices) were recruited to deliver a medical guidewire to two target arteries on a porcine model for in vivo studies. Simultaneously, translation and twist manipulations of thumb, forefinger, and wrist are acquired with electromagnetic (EM) and fiber-optic bend (FOB) sensors, respectively. These behavior data are then processed with wavelet packet decomposition (WPD) under 1-10 levels for feature extraction. The feature vectors are further fed into three candidate individual classifiers in the local learning layer. Furthermore, the local learning results from different manipulation behaviors are fused in the ensemble learning layer with three rule-based ensemble learning algorithms. In subject-dependent skill characteristics learning, the ensemble learning can achieve 100% accuracy, significantly outperforming the best local result (90%). Furthermore, ensemble learning can also maintain 73% accuracy in subject-independent schemes. These promising results demonstrate the great potential of the proposed method to facilitate skill learning in surgical robotics and skill assessment in clinical practice.
7 January 2022 in IEEE Transactions on Medical Robotics and Bionics
Percutaneous coronary intervention (PCI) has become a popular treatment for coronary artery disease. Highly dexterous skills are necessary to procedure success. However, few effective methods can be applied to PCI skill assessment. In this study, ten interventional cardiologists (four experts and six novices) were recruited. In vivo studies were performed via delivering a medical guidewire into distal left circumflex artery (target vessel I) and obtuse marginal artery (target vessel II) of a porcine model. Regarded as a type of manipulation data, the guidewire motion is simultaneously acquired with an electromagnetic (EM) sensor attached to guidewire tail. To address the deficiency of conventional dynamic time warping (DTW) limited to two-sequence matching, a novel warping algorithm is proposed to match multiple manipulation data. Then the intra-similarity is calculated to evaluate consistencies among each subject’s different manipulations, while the inter-similarity is further analyzed to find skill differences among different subjects. Extensive statistical analysis demonstrates that the proposed algorithm can effectively distinguish between the manipulations made by different skill-level subjects with significant differences on the target vessel I ( $P = 3.25\times 10^{-4}$ ) and II ( $P = 7.30\times 10^{-3}$ ). These promising results show the proposed technique’s great potential to facilitate skill assessment in clinical practice and skill learning in surgical robotics.
31 January 2022 in IEEE Transactions on Instrumentation and Measurement
Robotic tactile sensing plays a vital role in realizing the precise perception of complex environments, affecting the application of robots in a wide range of industrial and household fields. However, existing tactile sensors are limited by weakly interpretable information mapping relationships. In this article, a novel Halbach-cylinder-based magnetic skin is constructed with permanent magnets and elastic rubber to address such problems. Furthermore, a highly interpretable and widely applicable method is deduced to effectively represent the skin’s information conversion as a Halbach cylinder deforms to an elliptical cylinder under external force. Experiments of simulated models and actual skins with different compressions are conducted to evaluate the proposed method. The results show that 0.24% and 0.26% relative errors in simulated and real environments, respectively, can be achieved. Our method can reach the minimum error in almost all situations compared with state-of-the-art approaches. These promising results demonstrate the method’s great potential to provide a basis for developing a practical magnetic-force mapping relationship.
30 May 2024 in IEEE Transactions on Medical Robotics and Bionics
Guidewire delivery is a fundamental procedure in percutaneous coronary intervention. The inherent flexibility of the guidewire poses challenges in precise control, necessitating long-term training and substantial expertise. In response, this paper proposes a novel offline reinforcement learning (RL) algorithm, Conservative Offline Reinforcement Learning with Variational Environment Model (CORVE), for autonomous delivery of guidewire. CORVE first uses offline data to train an environment model and then optimizes the policy with both offline and model-generated data. The proposed method shares an encoder between the environmental model, policy, and Qfunction, mitigating the common sample inefficiency in imagebased RL. Besides, CORVE utilizes model prediction errors to forecast wrong deliveries in inference, which is an attribute absent in existing methods. The experimental results show that CORVE obtains superior performance in guidewire deliveries, achieving notably higher success rates and smoother movements than existing methods. These findings suggest that CORVE holds significant potential for enhancing the autonomy of vascular robotic systems in clinical settings.
4 January 2023 in IEEE Transactions on Industrial Electronics
Soft magnetic skins have been widely adopted for tactile perception due to their high accuracy and simple wiring advantages. However, the perceptual properties of magnetic skins are limited by information mapping relationships with weak interpretation. To overcome existing limitations, dynamic Young's modulus (DYM) is proposed in this paper based on strain energy density function to precisely describe the compression stiffness of magnetic skins. Furthermore, a highly interpretable and broadly applicable method is derived using DYM to analyze a cylindrical magnetic skin's deformation process as the skin deformed under external mechanic load. Extensive experiments in simulated and real situations with different deformations are carried out to verify the proposed method. Experimental results demonstrate that 0.14% and 0.43% relative errors in simulated and real environments, respectively, can be reached. Moreover, the proposed method can achieve minimum errors in almost all situations than data-driven or state-of-the-art analysis approaches. And the generality of the proposed method is validated by experiments conducted on skins with two different shapes. These promising results indicate the potential of the proposed method in establishing practical information mapping relationships for magnetic skins, probably addressing the significant challenges for magnetic skin application in complex scenarios.
1 January 2025 in IEEE Transactions on Neural Systems and Rehabilitation Engineering
Abnormal gait recognition plays a critical role in healthcare, particularly for the early diagnosis and continuous monitoring of neurological and musculoskeletal disorders, such as Parkinson's disease and orthopedic injuries. This study proposes MSCAF-Gait, a Multi-Scale Cross-Attention Fusion Network designed specifically for abnormal gait recognition using foot pressure sensors. MSCAF-Gait incorporates multi-scale convolutional modules with channel and spatial attention mechanisms to effectively capture features across temporal, channel, and spatial dimensions. A novel cross-attention fusion module further enhances feature representation, enabling precise recognition of diverse abnormal gait patterns. To facilitate this research, we introduce the Pressure-Insole Abnormal Gait (PIAG) dataset, comprising gait data associated with common neurological and musculoskeletal abnormalities. Extensive experiments on the publicly available Gait in Parkinson's Disease (GaitinPD) dataset and our self-constructed PIAG dataset validate the effectiveness of MSCAF-Gait. Specifically, the model achieves 99.61% accuracy in Parkinsonian gait recognition and 98.88% accuracy in Parkinson's severity classification. On the PIAG dataset, which includes multiple abnormal gait patterns, MSCAF-Gait attains a high accuracy of 99.42%. Notably, these results are obtained with a lightweight architecture characterized by reduced FLOPs and parameter count, demonstrating that MSCAF-Gait offers both high accuracy and computational efficiency, making it well-suited for real-time deployment on wearable platforms.