GeMix: Conditional GAN-Based Mixup for Improved Medical Image Augmentation

2025-07-21

AI Summary

Traditional Mixup degrades robustness and clinical reliability in medical image classification due to pixel-level interpolation that produces semantically distorted samples. To address this, we propose a two-stage class-conditional GAN-based mixing augmentation framework. In Stage I, label-aware continuous class manifold interpolation is achieved using a pretrained StyleGAN2-ADA generator guided by soft labels. In Stage II, Dirichlet-prior sampling combined with Beta-distributed mixing coefficients ensures high-fidelity, visually coherent, and semantically plausible synthetic images. The method is plug-and-play—requiring no modification to backbone architectures or training pipelines. Evaluated on the COVIDx-CT-3 dataset across multiple backbones, it consistently improves macro-F1 by an average of +2.1% and reduces false-negative rate by up to 37%, demonstrating strong efficacy and generalizability for high-stakes medical diagnosis tasks.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Mixture of Experts (MoE)
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Mixup has become a popular augmentation strategy for image classification, yet its naive pixel-wise interpolation often produces unrealistic images that can hinder learning, particularly in high-stakes medical applications. We propose GeMix, a two-stage framework that replaces heuristic blending with a learned, label-aware interpolation powered by class-conditional GANs. First, a StyleGAN2-ADA generator is trained on the target dataset. During augmentation, we sample two label vectors from Dirichlet priors biased toward different classes and blend them via a Beta-distributed coefficient. Then, we condition the generator on this soft label to synthesize visually coherent images that lie along a continuous class manifold. We benchmark GeMix on the large-scale COVIDx-CT-3 dataset using three backbones (ResNet-50, ResNet-101, EfficientNet-B0). When combined with real data, our method increases macro-F1 over traditional mixup for all backbones, reducing the false negative rate for COVID-19 detection. GeMix is thus a drop-in replacement for pixel-space mixup, delivering stronger regularization and greater semantic fidelity, without disrupting existing training pipelines. We publicly release our code at https://github.com/hugocarlesso/GeMix to foster reproducibility and further research.

Problem

• Improves unrealistic medical images from naive mixup
• Uses GANs for label-aware image interpolation
• Reduces false negatives in COVID-19 detection

Innovation

• Uses class-conditional GANs for label-aware interpolation
• Combines StyleGAN2-ADA with Dirichlet and Beta sampling
• Enhances COVID-19 detection with semantic fidelity