DEVELOPMENT OF FEDERATED LYMPHOMA CLASSIFICATION MODELS ACROSS MULTIPLE HARMONIZED COHORTS OF PATIENTS WITH PRIMARY SJÖGREN’S SYNDROME

Abstract

Primary Sjögren’s Syndrome (pSS) is a chronic autoimmune disease followed by exocrine gland dysfunction, where it has been long stated that 5% of pSS patients are prone to lymphoma development. In this work, we present a federated AI (artificial intelligence) strategy which enables the federated training and validation of AI algorithms for lymphoma classification across 21 European cohorts with pSS patients. Advanced AI algorithms were developed, including federated gradient boosting trees with and without dropouts, federated Multilayer Perceptron and federated Multinomial Naïve Bayes. Two large-scale case studies were conducted to demonstrate the applicability and robustness of the federated AI models, where emphasis is given on class imbalance handling and explainability analysis. The federated gradient boosting trees with dropouts achieved the best classification performance yielding more than 0.8 sensitivity and specificity along with 5 biomarkers as prominent for lymphoma development and progression.