R3GAN-based Optimal Strategy for Augmenting Small Medical Dataset

2025-10-29

AI Summary

To address data scarcity and severe class imbalance hindering deep learning’s clinical deployment in medical imaging, this paper proposes R3GAN—an optimized generative framework tailored for few-shot scenarios, using human embryo time-lapse imaging (TLI) as a case study. The method introduces a novel training strategy featuring a comprehensive warm-up phase and progressive gamma adjustment, integrated with systematic hyperparameter optimization, lightweight data augmentation, and explicit class-balancing mechanisms. R3GAN synthesizes high-fidelity, diagnostically meaningful embryo images at 256×256 resolution. Experiments demonstrate substantial improvement on the highly imbalanced t3 class: recall increases from 0.06 to 0.69, and F1-score rises from 0.11 to 0.60, while performance on other classes remains stable. These results validate both the clinical utility of the generated images and the model’s robust generalization capability under extreme label skew.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Medical image analysis often suffers from data scarcity and class imbalance, limiting the effectiveness of deep learning models in clinical applications. Using human embryo time-lapse imaging (TLI) as a case study, this work investigates how generative adversarial networks (GANs) can be optimized for small datasets to generate realistic and diagnostically meaningful images. Based on systematic experiments with R3GAN, we established effective training strategies and designed an optimized configuration for 256x256-resolution datasets, featuring a full burn-in phase and a low, gradually increasing gamma range (5 -> 40). The generated samples were used to balance an imbalanced embryo dataset, leading to substantial improvement in classification performance. The recall and F1-score of t3 increased from 0.06 to 0.69 and 0.11 to 0.60, respectively, without compromising other classes. These results demonstrate that tailored R3GAN training strategies can effectively alleviate data scarcity and improve model robustness in small-scale medical imaging tasks.

Problem

• Optimizing GANs for small medical datasets
• Addressing data scarcity and class imbalance
• Improving classification with generated embryo images

Innovation

• R3GAN generates realistic medical images from small datasets
• Optimized training uses full burn-in and gradual gamma increase
• Balances imbalanced datasets to improve classification performance