Pub Date : 2025-10-06DOI: 10.1109/TMRB.2025.3617956
Navid Feizi;Filipe C. Pedrosa;Ruisi Zhang;Dianne Sacco;Rajni V. Patel;Jagadeesan Jayender
Concentric-tube robots (CTRs) have garnered significant attention in various minimally invasive procedures due to their small size and dexterity. Despite extensive technical advancements in the development of CTRs, there is a lack of design approaches specific to their function as surgical instruments. This study proposes a compact CTR specifically designed for percutaneous nephrolithotomy (PCNL), adaptable for both hand-held operation and mounting on a passive arm. We employ a parallel carriage-based design to reduce the device’s cross-sectional footprint (46 mm diameter, 322 mm length) and localize the center of mass (570 g mass) beneath the grip area, enhancing ergonomic comfort and control. An ergonomic evaluation of the robot during the handling of the robot by expert urologists, as well as non-clinicians, showed better ergonomics than standard hand-held PCNL devices. Additionally, closed-loop position control of the distal end of the CTR was implemented based on resolved-motion rate inverse kinematics. The performance of the robot was empirically validated through experiments on a life-size abdominal phantom. The results showed mean closed-loop position errors of 1.20.8 mm for autonomous navigation to 100 target points on the stone, indicating a performance level in line with the specific requirements of PCNL.
{"title":"Design and Validation of a Compact Concentric-Tube Robot for Percutaneous Nephrolithotomy","authors":"Navid Feizi;Filipe C. Pedrosa;Ruisi Zhang;Dianne Sacco;Rajni V. Patel;Jagadeesan Jayender","doi":"10.1109/TMRB.2025.3617956","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3617956","url":null,"abstract":"Concentric-tube robots (CTRs) have garnered significant attention in various minimally invasive procedures due to their small size and dexterity. Despite extensive technical advancements in the development of CTRs, there is a lack of design approaches specific to their function as surgical instruments. This study proposes a compact CTR specifically designed for percutaneous nephrolithotomy (PCNL), adaptable for both hand-held operation and mounting on a passive arm. We employ a parallel carriage-based design to reduce the device’s cross-sectional footprint (46 mm diameter, 322 mm length) and localize the center of mass (570 g mass) beneath the grip area, enhancing ergonomic comfort and control. An ergonomic evaluation of the robot during the handling of the robot by expert urologists, as well as non-clinicians, showed better ergonomics than standard hand-held PCNL devices. Additionally, closed-loop position control of the distal end of the CTR was implemented based on resolved-motion rate inverse kinematics. The performance of the robot was empirically validated through experiments on a life-size abdominal phantom. The results showed mean closed-loop position errors of 1.20.8 mm for autonomous navigation to 100 target points on the stone, indicating a performance level in line with the specific requirements of PCNL.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1739-1754"},"PeriodicalIF":3.8,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-06DOI: 10.1109/TMRB.2025.3617959
Dezhi Sun;Stefano Stramigioli;Kenan Niu
Accurate segmentation of musculoskeletal structures in ultrasound (US) images remains challenging due to speckle noise, multi-layer anatomical boundaries, and scanning variability. For the robotic ultrasound system, the quality of captured ultrasound images highly depends on the force and angle applied to the tissue during autonomous scanning. Consequently, how the autonomous scan is performed influences the subsequent image segmentation task. Particularly, segmentation algorithms for bone structures are relatively less affected by variations in applied force. In contrast, muscle segmentation remains particularly challenging due to tissue deformation caused by variations in applied force during robotic scanning. Existing algorithms typically focus on either bone or muscle, rather than addressing both structures simultaneously. To address those challenges, we proposed an autonomous robotic ultrasound system that integrates precise force control with a cascaded deep learning framework in this paper. Specifically, the Hybrid Channel and Coordinate Enhanced Cascaded U-Net (HCCE-CUNet) was designed to enable simultaneous segmentation of bone and multi-layer muscle structures with improved accuracy. Experimental evaluations on two customized forearm phantoms demonstrated the system’s reliability, achieving a root-mean-square error in force tracking below 0.14N, and showed significant segmentation improvements, with Dice coefficients of 0.8915 (single-layer phantom) and 0.9175 (multi-layer phantom). The proposed segmentation method extends the image processing capability of the robotic ultrasound to deal with hard tissues (i.e., bones) and multiple muscles simultaneously. In the future, it could have great potential to provide a reliable solution for operator-independent musculoskeletal diagnostics and interventions.
{"title":"HCCE-CUNet-Based Multi-Class Musculoskeletal Segmentation for Robotic Ultrasound System","authors":"Dezhi Sun;Stefano Stramigioli;Kenan Niu","doi":"10.1109/TMRB.2025.3617959","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3617959","url":null,"abstract":"Accurate segmentation of musculoskeletal structures in ultrasound (US) images remains challenging due to speckle noise, multi-layer anatomical boundaries, and scanning variability. For the robotic ultrasound system, the quality of captured ultrasound images highly depends on the force and angle applied to the tissue during autonomous scanning. Consequently, how the autonomous scan is performed influences the subsequent image segmentation task. Particularly, segmentation algorithms for bone structures are relatively less affected by variations in applied force. In contrast, muscle segmentation remains particularly challenging due to tissue deformation caused by variations in applied force during robotic scanning. Existing algorithms typically focus on either bone or muscle, rather than addressing both structures simultaneously. To address those challenges, we proposed an autonomous robotic ultrasound system that integrates precise force control with a cascaded deep learning framework in this paper. Specifically, the Hybrid Channel and Coordinate Enhanced Cascaded U-Net (HCCE-CUNet) was designed to enable simultaneous segmentation of bone and multi-layer muscle structures with improved accuracy. Experimental evaluations on two customized forearm phantoms demonstrated the system’s reliability, achieving a root-mean-square error in force tracking below 0.14N, and showed significant segmentation improvements, with Dice coefficients of 0.8915 (single-layer phantom) and 0.9175 (multi-layer phantom). The proposed segmentation method extends the image processing capability of the robotic ultrasound to deal with hard tissues (i.e., bones) and multiple muscles simultaneously. In the future, it could have great potential to provide a reliable solution for operator-independent musculoskeletal diagnostics and interventions.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1728-1738"},"PeriodicalIF":3.8,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659193","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1109/TMRB.2025.3604121
Marco Moran-Ledesma;Robert Burns;Mark Hancock;Oliver Schneider
We present the design and implementation of a high-fidelity haptic manikin for knee injury assessment training. Currently, such training is conducted through direct instruction on live patients or peer-to-peer practice, which may limit exposure to multiple injury severities and raise ethical concerns. Our manikin aims to assist inexperienced practitioners in mastering an injury assessment technique specifically for the medial collateral ligament (MCL). We designed the manikin collaboratively with a certified clinician. Our design incorporates a commercial human knee joint model for accurate anatomical representation, materials that closely mimic human skin properties, an injury simulation mechanism for replicating MCL injuries, and pressure sensors to capture user-applied pressure during manipulation. We conducted three evaluations: an internal test with our collaborating clinician to configure our manikin for four MCL injury conditions (i.e., healthy, grade 1, grade 2, and grade 3) using a psychophysics method; a subsequent study where 6 certified clinicians rated each condition for consistency and a technical evaluation measuring abduction range in the healthy and grade 3 configurations. Results show that our manikin can reliably display healthy and unhealthy MCLs However, further improvements are needed to accurately distinguish between injury grades. Our manikin’s realistic weight and shape were highly praised, but there is room for improvement in simulating the skin texture. This work shows the potential of realistic simulators to enhance clinical training with standardized and repeatable practice.
{"title":"TRAIN-KNEE: Developing a Haptic Manikin for Knee Injury Assessment Training","authors":"Marco Moran-Ledesma;Robert Burns;Mark Hancock;Oliver Schneider","doi":"10.1109/TMRB.2025.3604121","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604121","url":null,"abstract":"We present the design and implementation of a high-fidelity haptic manikin for knee injury assessment training. Currently, such training is conducted through direct instruction on live patients or peer-to-peer practice, which may limit exposure to multiple injury severities and raise ethical concerns. Our manikin aims to assist inexperienced practitioners in mastering an injury assessment technique specifically for the medial collateral ligament (MCL). We designed the manikin collaboratively with a certified clinician. Our design incorporates a commercial human knee joint model for accurate anatomical representation, materials that closely mimic human skin properties, an injury simulation mechanism for replicating MCL injuries, and pressure sensors to capture user-applied pressure during manipulation. We conducted three evaluations: an internal test with our collaborating clinician to configure our manikin for four MCL injury conditions (i.e., healthy, grade 1, grade 2, and grade 3) using a psychophysics method; a subsequent study where 6 certified clinicians rated each condition for consistency and a technical evaluation measuring abduction range in the healthy and grade 3 configurations. Results show that our manikin can reliably display healthy and unhealthy MCLs However, further improvements are needed to accurately distinguish between injury grades. Our manikin’s realistic weight and shape were highly praised, but there is room for improvement in simulating the skin texture. This work shows the potential of realistic simulators to enhance clinical training with standardized and repeatable practice.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1777-1788"},"PeriodicalIF":3.8,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659208","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Robot-assisted intervention at a higher level of autonomy can reduce the surgical complexity and physician workload. However, as the core requirement, the autonomous delivery of the guidewire remains challenging. The inherent flexibility of the guidewire complicates physical modeling, while existing learning-based approaches require prolonged training and exhibit limited interpretability. To address these issues, this paper proposes a novel method, experience-based Fuzzy LOgic framework for Robot-assisted endovascular Intervention (FLORI). FLORI emulates clinical operator techniques by leveraging a fuzzy logic system. It dynamically computes control parameters by analyzing both real-time guidewire tip positioning within the vascular architecture and historical attempt data. The proposed method maintains interpretability while improving the success rate and efficiency of guidewire delivery. Furthermore, this paper introduces a “macro + micro” human-machine collaborative delivery paradigm that allows physicians to switch between autonomous and manual delivery modes, as well as modify the guidewire delivery path during the surgical procedure. Physical experiments on a coronary artery model have demonstrated the effectiveness of FLORI and the role of the human-machine collaborative paradigm in reducing physician workload and enhancing the success rate of surgical procedures.
{"title":"Experience-Based Fuzzy Logic Framework for Robot-Assisted Endovascular Intervention","authors":"Pei-Liang Wu;Ding-Qian Wang;Xiao-Hu Zhou;Mei-Jiang Gui;Xiao-Liang Xie;Shi-Qi Liu;Shuang-Yi Wang;Hao Li;Zhi-Chao Lai;Zeng-Guang Hou","doi":"10.1109/TMRB.2025.3604114","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604114","url":null,"abstract":"Robot-assisted intervention at a higher level of autonomy can reduce the surgical complexity and physician workload. However, as the core requirement, the autonomous delivery of the guidewire remains challenging. The inherent flexibility of the guidewire complicates physical modeling, while existing learning-based approaches require prolonged training and exhibit limited interpretability. To address these issues, this paper proposes a novel method, experience-based Fuzzy LOgic framework for Robot-assisted endovascular Intervention (FLORI). FLORI emulates clinical operator techniques by leveraging a fuzzy logic system. It dynamically computes control parameters by analyzing both real-time guidewire tip positioning within the vascular architecture and historical attempt data. The proposed method maintains interpretability while improving the success rate and efficiency of guidewire delivery. Furthermore, this paper introduces a “macro + micro” human-machine collaborative delivery paradigm that allows physicians to switch between autonomous and manual delivery modes, as well as modify the guidewire delivery path during the surgical procedure. Physical experiments on a coronary artery model have demonstrated the effectiveness of FLORI and the role of the human-machine collaborative paradigm in reducing physician workload and enhancing the success rate of surgical procedures.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1715-1727"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659201","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
While sliding a finger over a surface, the perceived hand position results from musculoskeletal proprioception, motor commands, and tactile motion estimate. When the touched surface has parallel raised ridges, tactile estimate is biased toward a direction perpendicular to the ridges, predicted by the tactile flow model. This illusory effect leads to systematic errors in the reaching movement that depends on ridge orientation and tactile sensitivity. This suggests the fascinating hypothesis that reaching tasks can be used for functional assessment of tactile deficit, a common symptom in several neurological diseases, and for therapeutic intervention. Previously, we demonstrated in simulations how this phenomenon can be used to guide the user’s finger sliding on a ridged plate to a target while the user is instructed to move toward another target. This is achieved by designing a Model Predictive Control strategy to estimate optimal ridge orientation at each time instant. In this study, we aim to replicate this behavior with a robotic manipulator endowed with a soft optical tactile sensor to detect surface ridges relying on deep learning techniques to estimate optical flow as tactile flow approximation. A biomimetic robotic architecture replicating in a controllable fashion such behavior represents a unique testbed for neuroscientific investigation and the design of subject-tailored rehabilitation protocols.
{"title":"Controlling Robot Sliding Relying on Tactile Sensors and Computational Models of Human Touch","authors":"Giulia Pagnanelli;Marco Greco;Paolo Susini;Alessandro Moscatelli;Matteo Bianchi","doi":"10.1109/TMRB.2025.3604093","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604093","url":null,"abstract":"While sliding a finger over a surface, the perceived hand position results from musculoskeletal proprioception, motor commands, and tactile motion estimate. When the touched surface has parallel raised ridges, tactile estimate is biased toward a direction perpendicular to the ridges, predicted by the tactile flow model. This illusory effect leads to systematic errors in the reaching movement that depends on ridge orientation and tactile sensitivity. This suggests the fascinating hypothesis that reaching tasks can be used for functional assessment of tactile deficit, a common symptom in several neurological diseases, and for therapeutic intervention. Previously, we demonstrated in simulations how this phenomenon can be used to guide the user’s finger sliding on a ridged plate to a target while the user is instructed to move toward another target. This is achieved by designing a Model Predictive Control strategy to estimate optimal ridge orientation at each time instant. In this study, we aim to replicate this behavior with a robotic manipulator endowed with a soft optical tactile sensor to detect surface ridges relying on deep learning techniques to estimate optical flow as tactile flow approximation. A biomimetic robotic architecture replicating in a controllable fashion such behavior represents a unique testbed for neuroscientific investigation and the design of subject-tailored rehabilitation protocols.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1755-1764"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659205","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-29DOI: 10.1109/TMRB.2025.3604139
Yingxin Qiu;Mengnan Wu;Lena H. Ting;Jun Ueda
System identification of human sensorimotor systems requires multiple experimental trials to achieve reliable parameter estimates, yet practical constraints limit the total number of trials possible. While pseudorandom sequence (PRS) perturbations are widely used due to their white noise-like properties, and optimal multisines can theoretically provide better performance when prior system knowledge is available, their implementation on mechanical devices presents significant challenges. Device dynamics can degrade the designed spectral properties of both perturbation types, increasing the number of required trials to achieve desired estimation precision. This paper presents a foundational framework for device-dynamics-aware perturbation design that reduces the necessary number of experimental trials. The framework introduces two key components: a prefilter for PRS to minimize digital-to-analog conversion effects, and a modified cost function for multisine optimization that explicitly compensates for mechanical device dynamics. We propose a two-stage approach where the prefiltered PRS first provides initial estimates that inform subsequent optimal multisine design. Through human arm impedance experiments and device-rendered validation, we demonstrate that our framework achieves much smaller covariance resulting in fewer trials to achieve satisfactory identification performance compared to conventional methods. The optimal multisine stage, enhanced by device dynamics compensation, shows particular effectiveness in reducing parameter covariance. The covariance improvement translates to multiple practical benefits: a potential 62.5% reduction in required trial numbers when full-length signals are used, a 75% reduction in single-trial duration while maintaining estimation quality, or various combinations of these improvements depending on experimental constraints. These results establish a practical path toward more efficient human system identification protocols that minimize experimental burden while maintaining estimation accuracy.
{"title":"Two-Stage Optimized Perturbation Design for Efficient Human Arm Impedance Identification With Device Dynamics Compensation","authors":"Yingxin Qiu;Mengnan Wu;Lena H. Ting;Jun Ueda","doi":"10.1109/TMRB.2025.3604139","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604139","url":null,"abstract":"System identification of human sensorimotor systems requires multiple experimental trials to achieve reliable parameter estimates, yet practical constraints limit the total number of trials possible. While pseudorandom sequence (PRS) perturbations are widely used due to their white noise-like properties, and optimal multisines can theoretically provide better performance when prior system knowledge is available, their implementation on mechanical devices presents significant challenges. Device dynamics can degrade the designed spectral properties of both perturbation types, increasing the number of required trials to achieve desired estimation precision. This paper presents a foundational framework for device-dynamics-aware perturbation design that reduces the necessary number of experimental trials. The framework introduces two key components: a prefilter for PRS to minimize digital-to-analog conversion effects, and a modified cost function for multisine optimization that explicitly compensates for mechanical device dynamics. We propose a two-stage approach where the prefiltered PRS first provides initial estimates that inform subsequent optimal multisine design. Through human arm impedance experiments and device-rendered validation, we demonstrate that our framework achieves much smaller covariance resulting in fewer trials to achieve satisfactory identification performance compared to conventional methods. The optimal multisine stage, enhanced by device dynamics compensation, shows particular effectiveness in reducing parameter covariance. The covariance improvement translates to multiple practical benefits: a potential 62.5% reduction in required trial numbers when full-length signals are used, a 75% reduction in single-trial duration while maintaining estimation quality, or various combinations of these improvements depending on experimental constraints. These results establish a practical path toward more efficient human system identification protocols that minimize experimental burden while maintaining estimation accuracy.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1658-1669"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659189","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-29DOI: 10.1109/TMRB.2025.3604146
Paniz Sedighi;Xingyu Li;Vivian K. Mushahwar;Mahdi Tavakoli
Personalization in the myoelectric control of robotic exoskeletons is crucial to ensuring accurate interpretation and adaptation to the unique muscle activity patterns and movement intentions of each user. This approach minimizes the risk of incorrect or excessive force application, significantly reducing the likelihood of user discomfort or injury during operation. This study introduces a model-agnostic meta-learning approach for personalizing a soft upper-limb exoskeleton in industrial settings. The framework incorporates an attention-based CNN-LSTM model that predicts future angular positions of the robot using EMG and IMU signals. The MAML framework demonstrates significant adaptability and personalization, efficiently predicting future angular positions with minimal data, approximately 20-25 seconds per task. This approach effectively reduces the necessity for extensive retraining with new users or in new environments by 50%, showcasing real-time task adaptation capabilities. Our findings confirmed a reduced human effort of nearly 13% in load-bearing tasks. Also, the results show that the exerted torque from the exoskeleton was 24% higher while maintaining higher accuracy. A comparison with other deep learning models further emphasizes the enhanced adaptability and accuracy offered by the meta-learning approach.
{"title":"Personalized Myoelectric Control for Upper-Limb Exoskeletons Through Meta-Learning: A Few-Shot Learning Approach","authors":"Paniz Sedighi;Xingyu Li;Vivian K. Mushahwar;Mahdi Tavakoli","doi":"10.1109/TMRB.2025.3604146","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604146","url":null,"abstract":"Personalization in the myoelectric control of robotic exoskeletons is crucial to ensuring accurate interpretation and adaptation to the unique muscle activity patterns and movement intentions of each user. This approach minimizes the risk of incorrect or excessive force application, significantly reducing the likelihood of user discomfort or injury during operation. This study introduces a model-agnostic meta-learning approach for personalizing a soft upper-limb exoskeleton in industrial settings. The framework incorporates an attention-based CNN-LSTM model that predicts future angular positions of the robot using EMG and IMU signals. The MAML framework demonstrates significant adaptability and personalization, efficiently predicting future angular positions with minimal data, approximately 20-25 seconds per task. This approach effectively reduces the necessity for extensive retraining with new users or in new environments by 50%, showcasing real-time task adaptation capabilities. Our findings confirmed a reduced human effort of nearly 13% in load-bearing tasks. Also, the results show that the exerted torque from the exoskeleton was 24% higher while maintaining higher accuracy. A comparison with other deep learning models further emphasizes the enhanced adaptability and accuracy offered by the meta-learning approach.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1670-1680"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659227","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Advancements in deep learning for surgical instrument segmentation have notably improved the proficiency, safety, and efficacy of minimally invasive robotic surgeries. The effectiveness of deep learning, however, is contingent upon the availability of large datasets for training, which are often associated with substantial annotation costs. Given the dynamic nature of surgical robots, scribble-based labeling emerges as a more viable and cost-effective alternative to traditional pixel-wise dense labeling. This paper introduces the Scribble-Supervised Surgical Robotic Segmentation Transformer (S4RoboFormer), designed to mitigate the challenges posed by resource-intensive annotations. S4RoboFormer incorporates a Vision Transformer (ViT)-based U-shaped segmentation network, enhanced with a specialized Weakly-Supervised Learning (WSL) strategy that comprises consistency training through (i) data-based perturbation using a data-mixed interpolation technique, and (ii) network-based perturbation via a self-ensembling strategy. This methodology promotes uniform predictions across different levels of perturbation under conditions of limited-signal supervision. S4RoboFormer outperforms existing state-of-the-art baseline WSL frameworks with both Convolutional Neural Network (CNN)- and ViT-based segmentation networks on a pre-processed public dataset. The code of S4RoboFormer, all baseline methods, pre-processed data, and scribble simulation algorithm are all made publicly available at https://github.com/ziyangwang007/CV-WSL-Robot.
{"title":"S4RoboFormer: Scribble-Supervised Surgical Robotic Segmentation Transformer via Augmented Consistency Training","authors":"Ziyang Wang;Tianxiang Chen;Zi Ye;Yiyuan Ge;Zhihao Chen;Jiabao Li;Yifan Zhao","doi":"10.1109/TMRB.2025.3604103","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604103","url":null,"abstract":"Advancements in deep learning for surgical instrument segmentation have notably improved the proficiency, safety, and efficacy of minimally invasive robotic surgeries. The effectiveness of deep learning, however, is contingent upon the availability of large datasets for training, which are often associated with substantial annotation costs. Given the dynamic nature of surgical robots, scribble-based labeling emerges as a more viable and cost-effective alternative to traditional pixel-wise dense labeling. This paper introduces the Scribble-Supervised Surgical Robotic Segmentation Transformer (S4RoboFormer), designed to mitigate the challenges posed by resource-intensive annotations. S4RoboFormer incorporates a Vision Transformer (ViT)-based U-shaped segmentation network, enhanced with a specialized Weakly-Supervised Learning (WSL) strategy that comprises consistency training through (i) data-based perturbation using a data-mixed interpolation technique, and (ii) network-based perturbation via a self-ensembling strategy. This methodology promotes uniform predictions across different levels of perturbation under conditions of limited-signal supervision. S4RoboFormer outperforms existing state-of-the-art baseline WSL frameworks with both Convolutional Neural Network (CNN)- and ViT-based segmentation networks on a pre-processed public dataset. The code of S4RoboFormer, all baseline methods, pre-processed data, and scribble simulation algorithm are all made publicly available at <uri>https://github.com/ziyangwang007/CV-WSL-Robot</uri>.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1789-1793"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659202","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Real-time mitigation of signal noise in neuromorphic systems is a critical requirement for developing reliable implantable bionic interfaces targeting neurological disorders. While prior hardware implementations of neuronal models on FPGA have prioritized efficiency through approximations of nonlinear dynamics, they often neglect the stochastic nature of biological noise. In this work, we present a hardware implementation capable of real-time detection and correction of transient noise events using two well-established algorithms, regardless of their timing or duration. These algorithms were validated on Hodgkin-Huxley and FitzHugh-Nagumo models and synthesized on FPGA, confirming their precision, robustness, and feasibility for real-time deployment. Beyond conventional noise suppression, the proposed system models baseline biological activity and autonomously regulates abnormal deviations, potentially reducing neural dysfunction in implanted bioelectronic devices. This approach provides a foundational step toward adaptive neurobionic systems for therapeutic applications, such as neuroprosthetics or implantable controllers for managing chronic neurological disorders.
{"title":"Implantable FPGA-Based Neuromorphic System for Real-Time Noise Detection and Correction in Neurobionic Applications","authors":"Milad Ghanbarpour;Muhammad Akmal Chaudhary;Maher Assaad;Gilda Ghanbarpour","doi":"10.1109/TMRB.2025.3604098","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604098","url":null,"abstract":"Real-time mitigation of signal noise in neuromorphic systems is a critical requirement for developing reliable implantable bionic interfaces targeting neurological disorders. While prior hardware implementations of neuronal models on FPGA have prioritized efficiency through approximations of nonlinear dynamics, they often neglect the stochastic nature of biological noise. In this work, we present a hardware implementation capable of real-time detection and correction of transient noise events using two well-established algorithms, regardless of their timing or duration. These algorithms were validated on Hodgkin-Huxley and FitzHugh-Nagumo models and synthesized on FPGA, confirming their precision, robustness, and feasibility for real-time deployment. Beyond conventional noise suppression, the proposed system models baseline biological activity and autonomously regulates abnormal deviations, potentially reducing neural dysfunction in implanted bioelectronic devices. This approach provides a foundational step toward adaptive neurobionic systems for therapeutic applications, such as neuroprosthetics or implantable controllers for managing chronic neurological disorders.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1765-1776"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659220","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-29DOI: 10.1109/TMRB.2025.3604135
Luís Moreira;Joana Figueiredo;Cristina P. Santos
Robotic assistive devices have been equipped with locomotion mode (LM) decoding tools to adapt their assistance according to the user’s locomotion needs. However, most of the LM decoding tools are insufficient to predict the upcoming LM in advance; and do not consider the typically slow speeds of neurologically impaired users. This study aims to address these shortcomings by introducing an LM decoding tool to predict, in real-time, four LMs (standing, level-ground walking, stair descent, and stair ascent) when using a robotic assistive device (SmartOs system) at slow speeds. Thigh and shank segment angles feeding a Long Short-Term Memory provided the highest performance (accuracies of 98.6% and 97.2%, for able-bodied and stroke subjects, respectively). An LM-driven position control strategy was developed to assist the users based on the decoded LM. Real-time evaluations were performed under different speeds (self-selected and controlled), scenarios (indoor and outdoor), and assistance modes (zero-torque and LM-driven position controls). The proposed LM decoding tool showed low computational loads ($1.58~pm ~0.42$ ms). The SmartOs system was able to predict and adapt its assistance by $295~pm ~170$ ms before the user entered the new LM. The proposed control strategy is a step towards LM-based assistance for stroke patients.
{"title":"Real-Time Control of Ankle Orthosis Assistance Using a Locomotion Mode Prediction Tool","authors":"Luís Moreira;Joana Figueiredo;Cristina P. Santos","doi":"10.1109/TMRB.2025.3604135","DOIUrl":"https://doi.org/10.1109/TMRB.2025.3604135","url":null,"abstract":"Robotic assistive devices have been equipped with locomotion mode (LM) decoding tools to adapt their assistance according to the user’s locomotion needs. However, most of the LM decoding tools are insufficient to predict the upcoming LM in advance; and do not consider the typically slow speeds of neurologically impaired users. This study aims to address these shortcomings by introducing an LM decoding tool to predict, in real-time, four LMs (standing, level-ground walking, stair descent, and stair ascent) when using a robotic assistive device (SmartOs system) at slow speeds. Thigh and shank segment angles feeding a Long Short-Term Memory provided the highest performance (accuracies of 98.6% and 97.2%, for able-bodied and stroke subjects, respectively). An LM-driven position control strategy was developed to assist the users based on the decoded LM. Real-time evaluations were performed under different speeds (self-selected and controlled), scenarios (indoor and outdoor), and assistance modes (zero-torque and LM-driven position controls). The proposed LM decoding tool showed low computational loads (<inline-formula> <tex-math>$1.58~pm ~0.42$ </tex-math></inline-formula> ms). The SmartOs system was able to predict and adapt its assistance by <inline-formula> <tex-math>$295~pm ~170$ </tex-math></inline-formula> ms before the user entered the new LM. The proposed control strategy is a step towards LM-based assistance for stroke patients.","PeriodicalId":73318,"journal":{"name":"IEEE transactions on medical robotics and bionics","volume":"7 4","pages":"1681-1692"},"PeriodicalIF":3.8,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659156","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：