Exploiting Test-Time Augmentation in Federated Learning for Brain Tumor MRI Classification

TL;DR

This work tackles privacy-preserving brain-tumor MRI classification in a Federated Learning (FL) setting with non-IID data across sites. It compares models trained on Original versus Preprocessed inputs and investigates Test-Time Augmentation (TTA) during inference to assess robustness. Key findings show that preserving native MRI characteristics yields better federated performance than aggressive preprocessing, and that TTA provides a statistically significant and robust boost, especially when combined with light preprocessing ($p<0.001$, Cohen’s $d=1.80$, $+0.45$ percentage points). Practically, the study recommends using TTA as the default inference strategy in FL medical imaging and highlights scenarios where light preprocessing can yield additional gains within budget constraints.

Abstract

Efficient brain tumor diagnosis is crucial for early treatment; however, it is challenging because of lesion variability and image complexity. We evaluated convolutional neural networks (CNNs) in a federated learning (FL) setting, comparing models trained on original versus preprocessed MRI images (resizing, grayscale conversion, normalization, filtering, and histogram equalization). Preprocessing alone yielded negligible gains; combined with test-time augmentation (TTA), it delivered consistent, statistically significant improvements in federated MRI classification (p<0.001). In practice, TTA should be the default inference strategy in FL-based medical imaging; when the computational budget permits, pairing TTA with light preprocessing provides additional reliable gains.

Figures (6)

• Figure 1: Proposed Method.
• Figure 2: Schematic of the ResNet 18 architecture with basic residual blocks and four stages of increasing channel width.
• Figure 3: Basic FL topology.
• Figure 4: Original MRI scan and TTA-generated variations.
• Figure 5: Performance metrics without TTA.