Project information
- Category: Federated Neural Style Transfer
- Data: Chaoyang Datasets
- Project date: December, 2022
- Project URL: https://doi.org/10.18280/ts.390620
Collaborative Colorectal Cancer Classification on Highly Class Imbalanced Data Setting via Federated Neural Style Transfer Based Data Augmentation
Deep learning algorithms have shown significant progress in computational pathology, but their high data requirements hinder multi-institutional collaborations. Federated learning addresses this challenge by enabling collaborative model training across sites without exposing private data, though imbalanced data distributions can reduce performance. This study proposes a federated version of the neural style transfer algorithm as a data augmentation method for the class-imbalanced Chaoyang colorectal cancer dataset. The approach improved ACC, F1 Score, and AUC by 22.07%, 42.51%, and 9.65%, respectively, using only 20 content images and 5 style images, while experiments with varying content and style numbers produced consistent results.