Computer Assisted Surgery最新文献

This study advances surgical phase recognition in arthroscopic procedures, specifically Anterior Cruciate Ligament (ACL) reconstruction, by introducing the first arthroscopy dataset and a novel transformer-based model. We establish a benchmark for arthroscopic surgical phase recognition by leveraging spatio-temporal features to address challenges such as limited field of view, occlusions, and visual distortions. We developed the ACL27 dataset, comprising 27 videos of ACL surgeries, each labeled with surgical phases. Our model employs a transformer-based architecture, utilizing temporal-aware frame-wise feature extraction through ResNet-50 and transformer layers. This approach integrates spatio-temporal features and introduces a Surgical Progress Index (SPI) to quantify surgery progression. The model's performance was evaluated using accuracy, precision, recall, and Jaccard Index on the ACL27 and Cholec80 datasets. The proposed model achieved an overall accuracy of 72.9% on the ACL27 dataset. On the Cholec80 dataset, the model achieved performance comparable to state-of-the-art methods, with an accuracy of 92.4%. The SPI demonstrated an output error of 10.6% and 9.8% on ACL27 and Cholec80 datasets, respectively, indicating reliable surgery progression estimation. This study introduces a significant advancement in surgical phase recognition for arthroscopy, providing a comprehensive dataset and robust transformer-based model. The results validate the model's effectiveness and generalizability, highlighting its potential to improve surgical training, real-time assistance, and operational efficiency in orthopedic surgery. The publicly available dataset and code will facilitate future research in this critical field. Word Count: 6490.

State-of-the-art computer- and robot-assisted surgery systems rely on intraoperative imaging technologies such as computed tomography and fluoroscopy to provide detailed 3D visualizations of patient anatomy. However, these methods expose both patients and clinicians to ionizing radiation. This study introduces a radiation-free approach for 3D spine reconstruction using RGB-D data. Inspired by the "mental map" surgeons form during procedures, we present SurgPointTransformer, a shape completion method that reconstructs unexposed spinal regions from sparse surface observations. The method begins with a vertebra segmentation step that extracts vertebra-level point clouds for subsequent shape completion. SurgPointTransformer then uses an attention mechanism to learn the relationship between visible surface features and the complete spine structure. The approach is evaluated on an ex vivo dataset comprising nine samples, with CT-derived data used as ground truth. SurgPointTransformer significantly outperforms state-of-the-art baselines, achieving a Chamfer distance of 5.39 mm, an F-score of 0.85, an Earth mover's distance of 11.00 and a signal-to-noise ratio of 22.90 dB. These results demonstrate the potential of our method to reconstruct 3D vertebral shapes without exposing patients to ionizing radiation. This work contributes to the advancement of computer-aided and robot-assisted surgery by enhancing system perception and intelligence.

The objective of this study is to evaluate the safety of using augmented reality (AR) in laparoscopic (adeno)myomectomy, defined as an increase in operating time shorter than 15 min. A total of 17 AR cases underwent laparoscopic myomectomy or adenomyomectomy with the use of AR and 17 controls without AR for the resection of (adeno)myomas. The non-inferiority assumption was defined by an operative overtime not exceeding 15 min, representing 10% of the typical operative time. The 17 AR cases were matched to 17 controls. The criteria used in matching the two groups were the type of lesions, the size and the placement. The mean operative time was 135 ± 39 min for AR cases and 149 ± 62 min for controls. The margin of non-inferiority was expressed as a difference in operative time of 15 min between the case and control groups. The mean difference observed between AR cases and controls was -14 min with 90% CI [-38.3;11.3] and was significantly lower than the non-inferiority margin of 15 min (p = 0.03). This negative time difference means that the operative time is shorter for the AR cases group. Intraoperative data revealed a volume of bleeding ≤200 mL in 82.3% of AR cases and in 75% of controls (p = 0.62). No intra or postoperative complications were reported in the groups. The use of augmented reality in laparoscopic (adeno)myomectomy does not introduce additional constraints for the surgeon. It appears to be safe for the patients, with an absence of additional adverse events and of significantly prolonged operative time.

To develop a DeepSurv model for predicting survival in pancreatic adenocarcinoma patients, evaluating the benefit of surgical versus non-surgical treatment across different stages, including stage IV subcategories. Clinical data were extracted from the SEER database (2000-2020). Patients were randomly divided into a model-building group and an experimental group. The DeepSurv model was trained and hyperparameter-optimized. Simulated paired data were created by switching treatment status. Predicted survival rates were compared using generalized estimating equations. SHAP values analyzed variable importance.The study included 16,068 patients. The final model achieved a C-index of 0.85. Surgical treatment yielded higher survival rates than non-surgical across all stages (p<0.001), though the benefit diminished in advanced stages. For stage IV, surgery improved survival in T1-3 and N0 stages (p<0.001) but not in T4 and N1. SHAP analysis ranked M stage as the most significant predictor of mortality, followed by T stage, overall stage, and surgical status. M1 metastasis was associated with a 14% increased mortality risk, while surgery reduced risk by 11%.Surgery reduces mortality across stages, with declining efficacy in advanced disease. For stage IV patients, surgery is beneficial except for those with T4 or N1 disease. Combining DeepSurv with SHAP analysis facilitates individualized prediction of surgical survival benefits.

Exploring the application of 5E teaching model combined with virtual endoscopic surgical simulation system in surgical teaching. Eighty-six students who received standardized residency training in the Department of General Surgery at the Second Hospital of Shanxi Medical University from September 2022 to June 2023 were selected as the research subjects. They were randomly divided into experimental and control groups, with 43 students in each group. The experimental group adopts the 5E teaching mode combined with a virtual endoscopic surgery simulation system for teaching. In contrast, the control group was taught using traditional teaching and a simple endoscopic simulation training box. A comparison was made between the evaluation results, self-evaluation, and teaching mode evaluation of the two groups. A t-test was performed on two sets of measurement data using SPSS 26.0 software. The theoretical test scores (t = 17.240, p = 0.000) and skill test scores (t = 21.335, p = 0.000) of students in the experimental group were higher than those in the control group. Compared to the control group, the experimental group showed significant improvement in operational skills (t = 3.557, p = 0.001), knowledge application (t = 4.936, p = 0.000), and overall performance (t = 2.999, p = 0.003) after training. The attitudes of students in the experimental group toward ability training (t = 3.818, p = 0.000), class order (t = 3.189, p = 0.002), teaching mode (t = 2.955, p = 0.004), and teaching level evaluation (t = 6.238, p = 0.000) were significantly higher than those in the control group. The virtual endoscopic surgery simulation system combined with the 5E teaching mode can significantly improve the theoretical knowledge and clinical practice skills of resident physicians in standardized training. Suggest applying it to clinical teaching.

Background: Severe acetabular bone loss in complex primary total hip arthroplasty (THA) with dysplasia and post-traumatic defects poses a formidable challenge. We describe a novel technique integrating a custom-designed, 3D-printed titanium augment with open-platform robotic assistance to reconstruct a Paprosky type IIIB acetabular defect.

Methods: An elderly patient with severe dysplasia and chronic Paprosky IIIB acetabular defect underwent complex primary THA utilizing a patient-specific titanium augment and a cementless cup. Preoperative planning employed CT imaging and 3D modeling to ensure a precise defect fit and optimal cup support. An open-platform robotic system facilitated accurate reaming and component impaction according to the surgical plan.

Results: Intraoperatively, the augment was anchored to the host bone with screws, enabling placement of the cementless cup in an optimal orientation under robotic guidance. The construct restored the hip center of rotation and provided primary stability. The procedure proceeded without intraoperative complications. Estimated blood loss and operative time were recorded. Postoperative imaging demonstrated well-fixed augment and cup with anatomically restored hip center. Operative time: 361 minutes; blood loss: 3200 mL. Early rehabilitation proceeded without incident.

Conclusion: This case demonstrates the feasibility of combining patient-specific implants with robotic-assisted techniques in complex primary THA. The approach supported stable reconstruction of a substantial acetabular defect and suggested potential for enhanced precision in implant positioning and favorable early postoperative trajectories. Nevertheless, due to the single-case nature and limited follow-up, findings should be interpreted cautiously, and longer-term outcomes in larger, diverse cohorts are needed to determine broader applicability and durability.

This retrospective observational study aimed to explore the associations of AI-assisted CT-quantified C4/C5 skeletal muscle and adipose tissue indices with postoperative cervical kyphosis and long-term functional outcomes in patients undergoing laminoplasty. Postoperative cervical kyphosis is a prevalent complication of laminoplasty with incompletely elucidated pathogenesis, and the role of neck muscle and adipose tissue in this complication lacks validation via standardized 3D quantification. We enrolled 114 patients with cervical spondylosis who underwent laminoplasty at Wuhan Union Hospital between 2018 and 2022, excluding those with severe comorbidities. Preoperative CT scans (obtained within 3 months before surgery) were processed using a ResU-Net model to quantify C4/C5 tissue indices. Statistical analyses (SPSS 27.0) included multivariate logistic regression, and receiver operating characteristic (ROC) curves with the Youden index were used to determine predictive thresholds. Postoperative kyphosis was diagnosed based on routine follow-up cervical X-rays. Multivariate logistic regression revealed that C4 and C5 subcutaneous fat volume (SFV) were independently associated with postoperative kyphosis (p<0.05). The combined model integrating tissue indices and clinical variables achieved area under the curve (AUC) values of 0.706 (C4 SFV) and 0.717 (C5 SFV) (p<0.05). AI-assisted CT-quantified C4/C5 SFV is correlated with postoperative cervical kyphosis. Integration of CT-derived tissue metrics and clinical indicators improves the prediction of laminoplasty outcomes, providing a data-driven foundation for optimizing surgical planning and postoperative rehabilitation in cervical spondylosis patients.

High-quality intraoperative video documentation is increasingly valued in surgery for its role in surgical evaluation, procedural archiving, and education. However, the comparative advantages of different recording methods have not been thoroughly examined. In this prospective, double-blinded study, 44 experienced orthopedic surgeons evaluated posterior total hip arthroplasty videos recorded using three techniques: a head-mounted camera, a light-handle-mounted camera, and an externally operated camera. All videos were captured by the same surgeon using standardized hardware and settings. Participants assessed video quality and educational value using a structured questionnaire. Data were analyzed using ANOVA and chi-square testing. The light-handle-mounted camera received the highest mean scores across all five evaluation domains, including visual clarity, image stability, and overall quality (mean scores ranging from 6.91 to 7.98). Repeated measures ANOVA confirmed statistically significant differences among the three camera techniques for all five questions (p = 0.022-0.043). Post hoc analysis revealed that the light-handle-mounted camera significantly outperformed the head-mounted system (p < 0.05 for all comparisons), while the external camera also demonstrated superiority over the head-mounted method. Chi-square testing showed a significant difference in educational suitability ratings (Question 6), with the light-handle-mounted system receiving the highest percentage of affirmative responses (79.5%) compared to the head-mounted (50.0%) and external cameras (31.8%) (p < 0.001). The light-handle-mounted system offered the most balanced solution, providing stable, high-quality recordings without disrupting sterility or workflow. While head-mounted and external methods have niche applications, their practical limitations reduce their suitability for routine documentation in procedures.

Percutaneous femoral arterial access is a fundamental procedure in minimally invasive vascular interventions. However, inadequate visualization of the femoral artery may lead to inaccurate puncture and complications, with reported incidence rates of 3 to 18%. This study proposes a three-dimensional (3D) image-guided navigation system designed to enhance real-time visualization of the target vessel and puncture site during femoral artery access. This system employed an Iterative Closest Point (ICP)-based point cloud algorithm to achieve spatial registration between image space and patient space. An improved ICP method is implemented to optimize surface point cloud alignment, providing higher efficiency and accuracy compared to conventional approaches. Validation experiments were conducted using a standard model and a human phantom. Registration and navigation accuracy were quantified using fiducial registration error (FRE) for spatial alignment, target registration error (TRE) for navigation accuracy, and distance error for puncture precision. The system achieved a FRE of 0.944 mm. On the standard model, the average distance error was 0.885 mm, and the TRE was 0.915 mm. On the human phantom, the average distance error is 0.967 mm, and the average TRE is 0.981 mm. These results confirm the feasibility and effectiveness of the proposed 3D navigation system in guiding femoral artery puncture. All error metrics were within clinically acceptable thresholds, suggesting potential for improved procedural safety and precision in percutaneous vascular interventions.

Achieving optimal implant position and orientation during total knee arthroplasty (TKA) is a pivotal factor in long-term survival. Computer-assisted navigation (CAN) has been recognized as a trusted technology that improves the accuracy and consistency of femoral and tibial bone cuts. Imageless CAN offers advantages over image-based CAN by reducing cost, radiation exposure, and time. The purpose of this study was to evaluate the accuracy of an imageless optical navigation system for TKA in a clinical setting. Forty-two consecutive patients who underwent primary TKA with CAN were retrospectively reviewed. Femoral and tibial component coronal alignment was assessed via post-operative radiographs by two independent reviewers and compared against coronal alignment angles from the CAN. The primary outcome was the mean absolute difference of femoral and tibial varus/valgus angles between radiograph and intra-operative device measurements. Bland-Altman plots were used to assess agreement between the methods and statistically analyze potential systematic bias. The mean absolute differences between navigation-guided cut measurements and post-operative radiographs were 1.16 ± 1.03° and 1.76 ± 1.38° for femoral and tibial alignment respectively. About 88% of coronal measurements were within ±3°, while 99% were within ±5°. Bland-Altman analysis demonstrated a bias between CAN and radiographic measurements with CAN values averaging 0.52° (95% CI: 0.11°-0.93°) less than their paired radiographic measurements. This study demonstrated the ability of an optical imageless navigation system to measure, on average, femoral and tibial coronal cuts to within 2.0° of post-operative radiographic measurements in a clinical setting.