Search

Due to data privacy constraints, data sharing among multiple clinical centers is restricted, which impedes the development of high performance deep learning models from multicenter collaboration. Naive weight transfer methods share intermediate model weights without raw data and hence can bypass data privacy restrictions. However, performance drops are typically observed when the model is transferred from one center to the next because of the forgetting problem. Incremental transfer learning, which combines peer-to-peer federated learning and domain incremental learning, can overcome the data privacy issue and meanwhile preserve model performance by using continual learning techniques. In this work, a conventional domain/task incremental learning framework is adapted for incremental transfer learning. A survey on the efficacy of prevalent regularization-based continual learning methods for multicenter collaboration is performed. The influences of data heterogeneity, classifier head setting, network optimizer, model initialization, center order, and weight transfer type have been investigated thoroughly. Our framework is publicly accessible to the research community for further development.

This work aims to explore the impact of multicenter data heterogeneity on deep learning brain metastases (BM) autosegmentation performance, and assess the efficacy of an incremental transfer learning technique, namely learning without forgetting (LWF), to improve model generalizability without sharing raw data. A total of six BM datasets from University Hospital Erlangen (UKER), University Hospital Zurich (USZ), Stanford, UCSF, New York University (NYU), and BraTS Challenge 2023 were used. First, the performance of the DeepMedic network for BM autosegmentation was established for exclusive single-center training and mixed multicenter training, respectively. Subsequently privacy-preserving bilateral collaboration was evaluated, where a pretrained model is shared to another center for further training using transfer learning (TL) either with or without LWF. For single-center training, average F1 scores of BM detection range from 0.625 (NYU) to 0.876 (UKER) on respective single-center test data. Mixed multicenter training notably improves F1 scores at Stanford and NYU, with negligible improvement at other centers. When the UKER pretrained model is applied to USZ, LWF achieves a higher average F1 score (0.839) than naive TL (0.570) and single-center training (0.688) on combined UKER and USZ test data. Naive TL improves sensitivity and contouring accuracy, but compromises precision. Conversely, LWF demonstrates commendable sensitivity, precision and contouring accuracy. When applied to Stanford, similar performance was observed. Data heterogeneity (e.g., variations in metastases density, spatial distribution, and image spatial resolution across centers) results in varying performance in BM autosegmentation, posing challenges to model generalizability. LWF is a promising approach to peer-to-peer privacy-preserving model training. • A multicenter study of deep learning brain metastases autosegmentation is performed. • Data heterogeneity influences the generalizability of deep learning models for brain metastases autosegmentation. • Naïve transfer learning (TL) and learning without forgetting (LWF) can promote multicenter collaboration without sharing raw data. • TL demonstrates notable strengths in achieving high detection sensitivity and contouring accuracy, at the expense of precision. • LWF achieves a commendable equilibrium between sensitivity and precision, offering a more balanced approach to multicenter model development. [ABSTRACT FROM AUTHOR]

In this article, we review the history and progress of a large multicenter research network pertaining to emergency medical services for children. We describe the history, organization, infrastructure, and research agenda of the Pediatric Emergency Care Applied Research Network (PECARN), and highlight some of the important accomplishments since its inception. We also describe the network's strategy to grow its research portfolio, train new investigators, and study how to translate new evidence into practice. This strategy ensures not only the sustainability of the network in the future, but the growth of research in emergency medical services for children in general.

In this article, we review the history and progress of a large multicenter research network pertaining to emergency medical services for children. We describe the history, organization, infrastructure, and research agenda of the Pediatric Emergency Care Applied Research Network (PECARN), and highlight some of the important accomplishments since its inception. We also describe the network’s strategy to grow its research portfolio, train new investigators, and study how to translate new evidence into practice. This strategy ensures not only the sustainability of the network in the future, but the growth of research in emergency medical services for children in general.

Background Compliance with enhanced recovery programmes (ERPs) in oesophago-gastric (OG) surgery is challenging due to high rates of complications. Higher compliance is associated with better outcomes; however compliance is difficult to assess due to variable reporting and lack of standardisation of descriptors within an ERP. This study aims to understand variability in ERPs across two OG cancer centres and analyse impact of cross-site engagement and collaborative quality improvement (QI) on ERP compliance. Methods The collaborative QI project consisted of workshops and meetings that involved presentations and assessment of variations across ERPs, with prospective recording of ERP compliance at each site and cross-site feedback of results. Compliance and outcomes for 2016 were compared with 2017 to assess impact of QI initiative. Results At site A, overall mobilisation improved from 25% to 28% (P  = 0.448) and oral intake from 19% to 56% (P  = 0.002). There were no improvements in NGT (65% to 54%) and drain removal (69% to 53%). At site B, mobilisation improved from 51% to 68% (P  = 0.097), NGT removal from 37% to 51% (P  = 0.196), and drain removal from 15% to 32% (P  = 0.412). There was no improvement in oral intake (47% to 34%). At site A, all complications reduced from 56% to 46% (P  = 0.401), pneumonia from 47% to 22% (P  = 0.024), and grade 3 complications from 22% to 15% (P  = 0.605). At site B, all complications reduced from 58% to 54% (P  = 0.697), pneumonia from 46% to 37% (P  = 0.353), and grade 3 complications from 23% to 20% (P  = 0.678). At site A, the median length of hospital stay reduced from 11 days to 9 days (P  = 0.024), with no improvement at site B (12 to 14 days). Conclusion Engagement between clinical teams across the two sites resulted in improvement in postoperative outcomes. Compliance with individual ERP elements was observed across a few domains. Challenges encountered in this project included a lack of standardised definitions of ERP descriptors, variability in measures and variations in the individual elements within each ERP. Disclosure All authors have declared no conflicts of interest. [ABSTRACT FROM AUTHOR]