Preservation of Patient Level Privacy: Federated Classification and Calibration Models
Abstract
With the launching of the Precision Medicine Initiative in the United States, by the National Institute of Health, and the emergence of a large volume of electronic health records, there are many opportunities to improve clinical decision support systems. A large number of samples are needed to build predictive models that have adequate discrimination and calibration. However, protecting patient privacy is also an important issue. Patient data are typically protected in localized silos, and consolidation of datasets from different healthcare systems is difficult. Federated learning allows the training of a global model by amassing intermediate calculations from localized medical systems. The knowledge learned from the data can be transferred and aggregated to achieve better performance than the one achieved by individual local models. Federated learning may help build better models, providing more accurate predictions. There are two types of measures to assess how well a model performs: discrimination and calibration. While most papers report discrimination measures, calibration has often been neglected but it is a critical metric for evaluation. In this dissertation, I show a novel way to build classifiers and calibration models in a federated manner. I also show how I can evaluate and improve model calibration in this manner. Federated modeling enables the accumulation of knowledge and information that are otherwise locked behind local medical systems.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Efficient explainable federated learning
- Machine learning for all : a methodology for choosing a federated learning approach
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.