Surgery · Medical Literature
Privacy-Preserving Surgical Video Analysis with Swarm Learning - Results from a Multinational Appendectomy Cohort.
Abstract
Progress in artificial intelligence (AI)-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. 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 evaluated our pipeline using a dataset of 397 laparoscopic appendectomy recordings from six international centers for two patient-level staging tasks: (1) laparoscopic grading of appendicitis and appendiceal perforation detection; and (2) histopathologic inflammation grading. We identified optimal modeling configurations (frame sampling rates and model architectures) using the binary perforation detection task, then compared Swarm Learning with single-center and centralized learning across the laparoscopic and histopathologic disease staging tasks. In addition, we surveyed participating centers to identify barriers to clinical implementation of our learning pipeline for surgical video analysis. For binary perforation detection, frame sampling at one frame per second and use of the SurgTempoNet architecture resulted in reliable classification performance, outperforming SurgFrameNet and Multiple Instance Learning. For both laparoscopic (area under the receiver operating curve [AUROC]: 0.818±0.092) and histopathologic disease staging (AUROC: 0.626±0.029), Swarm Learning consistently outperformed single-center training and achieved performance comparable to centralized learning on external validation (AUROC: 0.795±0.092 for laparoscopic grading; AUROC: 0.610±0.018 for histopathologic grading). The user survey identified hardware failure and limited integration of the decentralized learning pipeline with electronic patient records as key barriers to clinical implementation. Weakly supervised deep learning enables the prediction of patient-level labels 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. (Funded by the European Union and others.).
Clinical relevance
This study addresses a critical bottleneck in surgical AI development: the inability to share sensitive laparoscopic video data across institutions due to privacy regulations, which has historically limited model generalizability and multicenter collaboration. The pipeline directly targets surgical patients undergoing laparoscopic appendectomy—a high-volume, globally performed procedure—by enabling outcome-level (patient-level) AI predictions without requiring labor-intensive frame-by-frame manual annotation. Clinicians and surgical informaticists should take note now as regulatory and institutional pressures around data sovereignty continue to intensify, making decentralized learning frameworks an increasingly necessary paradigm for real-world AI deployment.
Key findings
1. The optimized pipeline using SurgTempoNet architecture with 1 frame-per-second sampling outperformed both SurgFrameNet and Multiple Instance Learning approaches for binary appendiceal perforation detection, establishing the preferred modeling configuration. 2. Swarm Learning achieved an AUROC of 0.818 ± 0.092 for laparoscopic appendicitis/perforation grading on internal validation across the 6-center, 397-recording dataset. 3. For histopathologic inflammation grading, Swarm Learning achieved an AUROC of 0.626 ± 0.029 on internal validation—a more challenging task reflecting the indirect relationship between intraoperative video and pathology-level endpoints. 4. On external validation, Swarm Learning achieved performance comparable to fully centralized learning: AUROC 0.795 ± 0.092 for laparoscopic grading and AUROC 0.610 ± 0.018 for histopathologic grading, without any raw video data leaving individual institutional nodes. 5. Swarm Learning consistently outperformed single-center training on both staging tasks, quantitatively demonstrating the benefit of decentralized multicenter collaboration over siloed institutional models. 6. A stakeholder survey across participating centers identified hardware failure and lack of integration with electronic health record (EHR) systems as the primary barriers to clinical implementation of the decentralized pipeline.
Practice implications
Surgical departments and hospital informatics teams considering AI-assisted intraoperative or postoperative decision support should prioritize Swarm Learning-compatible infrastructure over data-pooling arrangements, as decentralized training achieves centralized-equivalent performance while satisfying data privacy requirements—particularly relevant under GDPR and analogous frameworks. For institutions participating in multicenter surgical AI consortia, investment in robust local hardware and EHR-pipeline integration (the two identified implementation barriers) should be treated as prerequisites before deployment, rather than afterthoughts. Clinically, these findings suggest that weakly supervised video-based AI could eventually augment surgeon-reported grading of appendicitis severity and perforation status in real time, potentially stratifying operative complexity and informing postoperative management without requiring dedicated annotation workflows.