Back

Federated Prediction Models and External Validation for Radiotherapy Outcomes in Oropharyngeal Cancer using F.A.I.R. Clinical and Radiomics Data

Gouthamchand, V.; Gottardelli, B.; Kulkarni, G.; Sherkhane, U. B.; Hogenboom, J.; Subramanian, R.; Jha, A. K.; Mithun, S.; Purandare, N. C.; Agarwal, J. P.; Sekar, K.; G, L.; Laskar, S. G.; Sinha, S.; Hoebers, F. J.; Rangarajan, V.; Sunder, G.; Dekker, A.; Soest, J. v.; Wee, L.

2025-01-29 health informatics
10.1101/2025.01.29.25321214 medRxiv
Show abstract

Development and validation of outcome prediction models in multi-centric cancer datasets are essential to ensure their applicability and accuracy across diverse populations. This study addresses the challenges of model generalizability in Head and Neck cancer research by utilizing combined clinical and radiomics data from centers in India and the Netherlands, following Findable-Accessible-Interoperable-Reusable (F.A.I.R.) data principles. We use Vantage6, a federated learning software infrastructure that implements the Personal Health Train (PHT) paradigm to ensure data privacy and security during collaborative research. Correlation-based Feature Selection (CFS) and LASSO regularized Cox regression were used to identify key features in training Cox proportional hazards models to predict Overall Survival (OS), Distant Metastasis (DM), and Locoregional Recurrence (LRR) in six datasets totaling 1131 oropharyngeal cancer patients. Our results highlight the efficacy of federated learning in providing a secure environment for multi-centric cancer research, enabling the development and validation of predictive models while upholding patient data confidentiality.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.