Pirmani Ashkan, De Brouwer Edward, Arany Ádám, Oldenhof Martijn, Passemiers Antoine, Faes Axel, Kalincik Tomas, Ozakbas Serkan, Gouider Riadh, Willekens Barbara, Horakova Dana, Havrdova Eva Kubala, Patti Francesco, Prat Alexandre, Lugaresi Alessandra, Tomassini Valentina, Grammond Pierre, Cartechini Elisabetta, Roos Izanne, Boz Cavit, Alroughani Raed, Amato Maria Pia, Buzzard Katherine, Lechner-Scott Jeannette, Guimarães Joana, Solaro Claudio, Gerlach Oliver, Soysal Aysun, Kuhle Jens, Sanchez-Menoyo Jose Luis, Spitaleri Daniele, Csepany Tunde, Van Wijmeersch Bart, Ampapa Radek, Prevost Julie, Khoury Samia J, Van Pesch Vincent, John Nevin, Maimone Davide, Weinstock-Guttman Bianca, Laureys Guy, McCombe Pamela, Blanco Yolanda, Altintas Ayse, Al-Asmi Abdullah, Garber Justin, Van der Walt Anneke, Butzkueven Helmut, de Gans Koen, Rozsa Csilla, Taylor Bruce, Al-Harbi Talal, Sas Attila, Rajda Cecilia, Gray Orla, Decoo Danny, Carroll William M, Kermode Allan G, Fabis-Pedrini Marzena, Mason Deborah, Perez-Sempere Angel, Simu Mihaela, Shuey Neil, Singhal Bhim, Cauchi Marija, Hardy Todd A, Ramanathan Sudarshini, Lalive Patrice, Sirbu Carmen-Adella, Hughes Stella, Castillo Trivino Tamara, Peeters Liesbet M, Moreau Yves
STADIUS, ESAT, KU Leuven, Leuven, Belgium.
Biomedical Research Institute, Hasselt University, Hasselt, Belgium.
NPJ Digit Med. 2025 Jul 24;8(1):478. doi: 10.1038/s41746-025-01788-8.
Early prediction of disability progression in multiple sclerosis (MS) remains challenging despite its critical importance for therapeutic decision-making. We present the first systematic evaluation of personalized federated learning (PFL) for 2-year MS disability progression prediction, leveraging multi-center real-world data from over 26,000 patients. While conventional federated learning (FL) enables privacy-aware collaborative modeling, it remains vulnerable to institutional data heterogeneity. PFL overcomes this challenge by adapting shared models to local data distributions without compromising privacy. We evaluated two personalization strategies: a novel AdaptiveDualBranchNet architecture with selective parameter sharing, and personalized fine-tuning of global models, benchmarked against centralized and client-specific approaches. Baseline FL underperformed relative to personalized methods, whereas personalization significantly improved performance, with personalized FedProx and FedAVG achieving ROC-AUC scores of 0.8398 ± 0.0019 and 0.8384 ± 0.0014, respectively. These findings establish personalization as critical for scalable, privacy-aware clinical prediction models and highlight its potential to inform earlier intervention strategies in MS and beyond.
尽管早期预测多发性硬化症（MS）的残疾进展对治疗决策至关重要，但仍然具有挑战性。我们利用来自26000多名患者的多中心真实世界数据，首次对用于预测MS两年残疾进展的个性化联邦学习（PFL）进行了系统评估。虽然传统的联邦学习（FL）能够实现隐私保护的协作建模，但它仍然容易受到机构数据异质性的影响。PFL通过使共享模型适应本地数据分布而不损害隐私，克服了这一挑战。我们评估了两种个性化策略：一种具有选择性参数共享的新型自适应双分支网络架构，以及对全局模型进行个性化微调，并与集中式方法和特定于客户端的方法进行了基准测试。相对于个性化方法，基线FL的表现较差，而个性化显著提高了性能，个性化联邦近端算法（FedProx）和联邦平均算法（FedAVG）的受试者工作特征曲线下面积（ROC-AUC）得分分别达到0.8398±0.0019和0.8384±0.0014。这些发现表明，个性化对于可扩展的、具有隐私保护意识的临床预测模型至关重要，并突出了其在为MS及其他疾病制定早期干预策略方面的潜力。
