Endoshare: a publicly available, surgeons-friendly solution to de-identify and manage surgical videos

Objectives: In this study we evaluated 40, top recommended, laparoscopic appendectomy and laparoscopic cholecystectomy videos located on public domain websites using eight criteria created by a panel of third year medical students and general surgeons. We hypothesized that there is a lack of quality, thorough educational laparoscopic surgical videos appropriate for third year medical students to review in preparation for the Surgery rotation. Methods: Utilizing a panel, which included four third year medical students and two general surgeons, we created an ‘ideal medical student educational video checklist.’ This checklist included 8 vital criteria. We selected 40, top recommended, videos available on YouTube and Google Video search engines, using ‘laparoscopic cholecystectomy’ and ‘laparoscopic appendectomy’ as key terms. Each video was evaluated by four third year medical students individually, using a binary system ‘meets’ or ‘does not meet’ each criterion. Individual scores were averaged, producing a single score for each video. Results: 0/40 (0%) of the videos met all eight of the criteria. 26/40 (65%) of the videos did not meet half of the criteria. The top performing videos 7/40 (17%) only met 5/8 criteria. Conclusions: We identified a lack of quality and thorough educational surgical videos appropriate for third year medical students and a need for improved online video based instruction. Our checklist can be utilized as a guide for anyone creating surgical videos for medical student education in the future.

Background:Progress in artificial intelligence-based analysis of surgical videos has been constrained by reliance on manual frame-level annotations rather than patient-level outcomes. In addition, concerns about data privacy restrict the exchange of laparoscopic video data and, thereby, multicenter collaboration.Methods:To address these limitations, we developed a pipeline that integrates weakly supervised deep learning with Swarm Learning, a decentralized machine learning approach that enables collaborative model training without data centralization. We evaluate our pipeline using a newly curated dataset of 397 laparoscopic appendectomy recordings from six international surgical centers. We identified optimal modelling configurations (frame sampling rates and model architectures) and subsequently compared Swarm Learning to single-center and centralized learning across three novel patient-level disease staging tasks: (i) binary detection of perforated appendicitis, (ii) laparoscopic grading of appendicitis, and (iii) histopathologic inflammation grading. In addition, we surveyed participating centers to identify real-world barriers to the clinical implementation of our decentralized learning pipeline for surgical video analysis.Results:For appendicitis grading tasks, frame sampling at 1.0 frames per second and use of the SurgTempoNet architecture resulted in reliable classification performance, outperforming SurgFrameNet and Multiple Instance Learning. Across all three disease staging tasks, Swarm Learning consistently outperformed single-center training and achieved performance comparable to centralized learning, with stable generalization in external validation. The user survey identified hardware failure and limited integration of the decentralized learning pipeline with electronic patient records as key barriers to the clinical implementation of our decentralized learning pipeline for collaborative surgical video analysis.Conclusions:Weakly supervised deep learning enables the prediction of patient-level endpoints directly from surgical video data. Swarm Learning facilitates privacy-preserving multicenter collaboration and achieves performance on par with centralized learning, highlighting its potential for advancing clinically relevant, collaborative AI development in surgical video analysis, especially when integrated with patients’ electronic health records.

Background:Laparoscopic D2 lymph node dissection (LND) for gastric cancer has a wide range and high difficulty. In the past, the quality of surgery was often judged by the time of the operation or the amount of blood loss, but the analysis based on surgical video was rarely reported. The purpose of this study was to analyze the relationship between the quality of laparoscopic D2 LND for gastric cancer and postoperative complications.Methods:The surgical video and clinicopathological data of 610 patients in two randomized controlled trials in our center from 2013 to 2016 were retrospectively analyzed. Klass-02-QC LND scale and general error score tool were used to quantitatively evaluate the intraoperative performance of D2 LND. Logistic regression was used to analyze the influencing factors of postoperative complications.Results:The overall incidence of complications (CD classification≥2) was 20.6%; the incidence of surgical complications was 6.9%. According to whether the LND score reached 44, patients were divided into a qualified group (73%) and a not-qualified group (27%). Event score (ES) by quartile was divided into grade 1 (21.7%), grade 2 (26%), grade 3 (28%), and grade 4 (24.3%) from low to high. Univariate logistic regression analysis showed that ES greater than or equal to 3, tumor size greater than or equal to 35 mm, and cTNM >II were independent risk factors for not-qualified LND. Male,tumor size greater than or equal to 35 mm and cTNM >II were independent risk factors for grade 4 ES. Not-qualified LND (OR=1.62, 95% CI: 1.16–3.89, P=0.021), grade 4 ES (OR=3.21, 95% CI: 1.52–3.90, P=0.035), and cTNM >II (OR=1.74, 95% CI: 1.39–7.33, P=0.041) were independent risk factors for postoperative surgical complications.Conclusions:The qualification of LND and intraoperative events based on surgical video are the independent influencing factors of postoperative complications of laparoscopic gastric cancer surgery. Specialist training and teaching based on surgical video may help to improve the surgical skills of specialists and improve the postoperative outcome of patients.

MP10-05 TIME TO OPTIMIZE UROLOGY TRAINEE EDUCATION AND PREPARATION FOR SURGERY: VIDEO-BASED SURGICAL LEARNING

INTRODUCTION: YouTube is a popular video search engine among surgical trainees due to its wide availability and easy accessibility. However, the educational quality of YouTube surgical videos may vary due to lack of requirement for peer review. In contrast, PubMed provides millions of peer-reviewed videos, but not all PubMed content is freely accessible. The Laparoscopic Surgery Video Educational Guidelines Video Assessment Tool (LAP-VEGaS VAT) is a nine-item checklist for assessing the educational quality of laparoscopic surgical videos. Each item is scored from 0 to 2, with higher scores indicating higher educational quality. To our knowledge, LAP-VEGaS VAT has not been used to evaluate sacrocolpopexy videos or to compare the quality of videos on PubMed versus YouTube. OBJECTIVE: To compare the educational quality of minimally invasive sacrocolpopexy surgical videos available on two platforms, PubMed and YouTube, using the LAP-VEGaS VAT METHODS: A PubMed search was performed for videos of minimally invasive sacrocolpopexy, applying keywords to capture relevant surgical methods. The “Video Audio Media” filter was used to isolate video content. Videos that were schematic, animated, included concomitant procedures, or contained marketing for surgical equipment were excluded. Each PubMed video was then matched to one or more YouTube videos based on minimally invasive route and video length. To reduce confounders to YouTube’s search algorithm, a private browser was utilized. The PubMed keywords were searched on YouTube, then matched from video list ranked by “Relevance.” Two surgeons (YC and SH) independently scored the videos using the LAP-VEGaS VAT. A third surgeon (MR), masked to initial scoring, resolved scoring discrepancies. The average cumulative and item-specific scores were compared between PubMed and YouTube using an independent-samples t-test, assuming unequal variances with corrections for multiple comparisons where applicable. RESULTS: 18 PubMed videos and 19 YouTube videos published between 2009 and 2024 were reviewed (Table 1). The overall median video length was 7 minutes 38 seconds (range, 4 minutes 30 seconds to 9 minutes 58 seconds). Traditional laparoscopy was the most common route of surgery (n=17, 45.9%), followed by robotic-assisted (n=13, 35.1%) and a combination transvaginal/laparoscopic approach (n=7, 18.9%). The mean cumulative LAP-VEGaS VAT score was significantly higher for the PubMed videos (13.2±2.9, range 9–17) than the YouTube videos (9.7±3.7, range 2–18, p=.003). PubMed videos scored on average higher across all individual items, especially for quality of formal case presentation (Q2), clarity of demonstration of intraoperative findings (Q5), and presence of English-language audio/written commentary (Q8) (Table 2). However, after correcting for multiple comparisons, these differences were not statistically significant. CONCLUSIONS: Educational surgical videos depicting minimally invasive sacrocolpopexy identified by PubMed search were of higher overall educational quality than those found on YouTube when evaluated with the LAP-VEGaS VAT. Restricted free access to many PubMed publications presents a potential barrier to surgeon ongoing medical education, particularly those working outside academic settings. Our findings highlight the importance of advocacy for increased access to publicly available peer-reviewed surgical content.Table 1Table 2

Background Optimizing the educational experience of trainees in the operating room is important; however, ear anatomy and otologic surgery are challenging for trainees to grasp. Viewing otologic surgeries often involves limitations related to video quality, such as visual disturbances and instability. Objectives We aimed to (1) improve the quality of surgical videos (tympanomastoidectomy [TM]) by using artificial intelligence (AI) techniques and (2) evaluate the effectiveness of processed videos through a questionnaire-based assessment from trainees. Materials and methods We conducted prospective study using video inpainting and stabilization techniques processed by AI. In each study set, we enrolled 21 trainees and asked them to watch processed videos and complete a questionnaire. Results Surgical videos with the video inpainting technique using the implicit neural representation (INR) model were found to be the most helpful for medical students (0.79 ± 0.58) in identifying bleeding focus. Videos with the stabilization technique via point feature matching were more helpful for low-grade residents (0.91 ± 0.12) and medical students (0.78 ± 0.35) in enhancing overall visibility and understanding surgical procedures. Conclusions and significance Surgical videos using video inpainting and stabilization techniques with AI were beneficial for educating trainees, especially participants with less anatomical knowledge and surgical experience.

Comparison of Robotic Surgery Video Quality Between YouTube and Curated Sources Using GEARS Criteria

To overcome limitations of open surgery artificial intelligence (AI) models by curating the largest collection of annotated videos and to leverage this AI-ready data set to develop a generalizable multitask AI model capable of real-time understanding of clinically significant surgical behaviors in prospectively collected real-world surgical videos. The study team programmatically queried open surgery procedures on YouTube and manually annotated selected videos to create the AI-ready data set used to train a multitask AI model for 2 proof-of-concept studies, one generating surgical signatures that define the patterns of a given procedure and the other identifying kinematics of hand motion that correlate with surgeon skill level and experience. The Annotated Videos of Open Surgery (AVOS) data set includes 1997 videos from 23 open-surgical procedure types uploaded to YouTube from 50 countries over the last 15 years. Prospectively recorded surgical videos were collected from a single tertiary care academic medical center. Deidentified videos were recorded of surgeons performing open surgical procedures and analyzed for correlation with surgical training. The multitask AI model was trained on the AI-ready video data set and then retrospectively applied to the prospectively collected video data set. Analysis of open surgical videos in near real-time, performance on AI-ready and prospectively collected videos, and quantification of surgeon skill. Using the AI-ready data set, the study team developed a multitask AI model capable of real-time understanding of surgical behaviors-the building blocks of procedural flow and surgeon skill-across space and time. Through principal component analysis, a single compound skill feature was identified, composed of a linear combination of kinematic hand attributes. This feature was a significant discriminator between experienced surgeons and surgical trainees across 101 prospectively collected surgical videos of 14 operators. For each unit increase in the compound feature value, the odds of the operator being an experienced surgeon were 3.6 times higher (95% CI, 1.67-7.62; P = .001). In this observational study, the AVOS-trained model was applied to analyze prospectively collected open surgical videos and identify kinematic descriptors of surgical skill related to efficiency of hand motion. The ability to provide AI-deduced insights into surgical structure and skill is valuable in optimizing surgical skill acquisition and ultimately improving surgical care.

Anchor-guided online meta adaptation for fast one-Shot instrument segmentation from robotic surgical videos.

What the Editors are reading: Social media

MP07-09 AUTOMATED SURGICAL STEP RECOGNITION IN TRANSURETHRAL BLADDER TUMOR RESECTION USING ARTIFICIAL INTELLIGENCE: TRANSFER LEARNING ACROSS SURGICAL MODALITIES

Background: Juvenile cystic adenomyosis (JCA) is a rare uterine lesion. We present the case of a young woman who was diagnosed with JCA and subsequently managed with laparoscopic cyst removal with sharp and blunt dissection. Moreover, we provide a literature review and a surgical video. Case: A 22-year-old nulliparous woman presented with severe dysmenorrhea and was assessed using contrast-enhanced abdominal computed tomography, transvaginal ultrasonography and pelvic magnetic resonance imaging, and diagnosed with a cystic lesion on the left side of the myometrium. She underwent laparoscopic cyst excision and uterine reconstruction. Histology was suggestive of JCA. The dysmenorrhea resolved postoperatively. Conclusion: Surgical resection is the first choice of treatment for cystic adenomyosis, and a laparoscopic approach using scissor forceps is effective.

Assessing nontechnical skills in operating rooms (ORs) is crucial for enhancing surgical performance and patient safety. However, automated and real-time evaluation of these skills remains challenging. To explore the feasibility of using motion features extracted from surgical video recordings to automatically assess nontechnical skills during cardiac surgical procedures. This cross-sectional study used video recordings of cardiac surgical procedures at a tertiary academic US hospital collected from January 2021 through May 2022. The OpenPose library was used to analyze videos to extract body pose estimations of team members and compute various team motion features. The Non-Technical Skills for Surgeons (NOTSS) assessment tool was employed for rating the OR team's nontechnical skills by 3 expert raters. NOTSS overall score, with motion features extracted from surgical videos as measures. A total of 30 complete cardiac surgery procedures were included: 26 (86.6%) were on-pump coronary artery bypass graft procedures and 4 (13.4%) were aortic valve replacement or repair procedures. All patients were male, and the mean (SD) age was 72 (6.3) years. All surgical teams were composed of 4 key roles (attending surgeon, attending anesthesiologist, primary perfusionist, and scrub nurse) with additional supporting roles. NOTSS scores correlated significantly with trajectory (r = 0.51, P = .005), acceleration (r = 0.48, P = .008), and entropy (r = -0.52, P = .004) of team displacement. Multiple linear regression, adjusted for patient factors, showed average team trajectory (adjusted R2 = 0.335; coefficient, 10.51 [95% CI, 8.81-12.21]; P = .004) and team displacement entropy (adjusted R2 = 0.304; coefficient, -12.64 [95% CI, -20.54 to -4.74]; P = .003) were associated with NOTSS scores. This study suggests a significant link between OR team movements and nontechnical skills ratings by NOTSS during cardiac surgical procedures, suggesting automated surgical video analysis could enhance nontechnical skills assessment. Further investigation across different hospitals and specialties is necessary to validate these findings.

Video is widely used in modern surgical education and may have advantages over other ways of learning surgery. However, less is known about the purposes and needs of surgical learners who use videos, as well as what learners consider to be the attributes of a ‘good’ surgical video. Qualitative research methods were used to answer these research questions. In-depth semi-structured interviews were conducted with nine medical students, eight surgical residents, and eight surgeons. The data were analyzed using a Thematic Analysis approach. Two frameworks were developed to structure the findings. Framework 1 explained that how learners used video was not simply associated with their designation but was related to the proficiency of an individual learner on a continuum of growing surgical proficiency from ‘seeing’ to ‘doing’ to ‘perfecting’. Framework 2 identified seven key attributes that learners associated with a ‘good’ surgical video: intelligible, concise, clear, interactive, reliable, accessible, and suitable. This study showed that a surgical video created for education requires to be tailored to the level of the surgical proficiency of the intended audience. The attributes of a ‘good’ surgical video need to be considered when deciding how to record, edit, and disseminate video-based educational materials in order to advance their value.

Surgical video analysis has become an essential component in computer-assisted interventions and clinical documentation. The rapid growth of minimally invasive surgery has produced large volumes of surgical recordings that require detailed frame-level annotations for training intelligent systems. Manual annotation of surgical videos remains a labor-intensive and time-consuming process that often requires expert knowledge. As a result, the development of automated annotation systems has become a critical research direction in medical image analysis. Existing segmentation and annotation approaches have faced limitations in handling complex surgical scenes, instrument occlusions, illumination variations, and tissue deformation. Conventional deep learning models often rely on large labelled datasets, whereas surgical datasets usually remain limited due to the difficulty of manual labeling. This challenge has reduced the reliability and scalability of automated surgical video segmentation systems. To address these issues, this study has proposed an Active Deep Ensemble Segmentation Network (ADES-Net) for automated surgical video segmentation annotation. The framework has integrated an ensemble of convolutional segmentation models with an active learning strategy that has selectively identified informative frames for annotation. The ensemble architecture has combined multiple deep segmentation networks that have captured diverse spatial representations from surgical frames. An uncertainty-driven active sampling mechanism has prioritized frames that required expert labeling, which has reduced redundant annotations. Feature representations that were extracted from each model have contributed to robust segmentation predictions, while iterative learning cycles have refined the annotation quality. The experimental evaluation demonstrates that the proposed ADES-Net framework achieves superior segmentation performance across multiple metrics. The model achieves a Dice similarity coefficient of 0.93, an IoU of 0.86, precision of 0.93, recall of 0.91, and an F1 score of 0.92 when trained with twenty-five annotated frames. These results indicate that the active ensemble mechanism effectively captures spatial and contextual features, reduces false positives, and improves boundary delineation. Compared with baseline methods such as U-Net, Attention U-Net, and DeepLabV3+, the proposed framework achieves improvements of 5–10% across all metrics, demonstrating enhanced segmentation reliability, efficiency, and robustness in automated surgical video annotation tasks.