Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MediAug: Exploring Visual Augmentation in Medical Imaging

Xuyin Qi, Zeyu Zhang, Canxuan Gang, Hao Zhang, Lei Zhang, Zhiwei Zhang, Yang Zhao

TL;DR

Medical image augmentation faces a domain gap with natural images and fragmented evaluation across tasks. MediAug provides a unified benchmark applying six mix-based augmentations—MixUp, YOCO, CropMix, CutMix, AugMix, and SnapMix—across ResNet-50 and ViT-B backbones on brain-tumor MRI and eye-fundus datasets, enabling direct cross-method comparisons. Results show task-specific best pairings (MixUp with ResNet-50 and SnapMix with ViT-B for brain tumors; YOCO with ResNet-50 and CutMix with ViT-B for eye diseases) and underscore the importance of hyperparameter tuning (e.g., CutMix $\alpha$). The framework, with public code, offers practical, reproducible guidance for robust, generalizable clinical AI systems and accelerates the safe deployment of deep learning in medical imaging.

Abstract

Data augmentation is essential in medical imaging for improving classification accuracy, lesion detection, and organ segmentation under limited data conditions. However, two significant challenges remain. First, a pronounced domain gap between natural photographs and medical images can distort critical disease features. Second, augmentation studies in medical imaging are fragmented and limited to single tasks or architectures, leaving the benefits of advanced mix-based strategies unclear. To address these challenges, we propose a unified evaluation framework with six mix-based augmentation methods integrated with both convolutional and transformer backbones on brain tumour MRI and eye disease fundus datasets. Our contributions are threefold. (1) We introduce MediAug, a comprehensive and reproducible benchmark for advanced data augmentation in medical imaging. (2) We systematically evaluate MixUp, YOCO, CropMix, CutMix, AugMix, and SnapMix with ResNet-50 and ViT-B backbones. (3) We demonstrate through extensive experiments that MixUp yields the greatest improvement on the brain tumor classification task for ResNet-50 with 79.19% accuracy and SnapMix yields the greatest improvement for ViT-B with 99.44% accuracy, and that YOCO yields the greatest improvement on the eye disease classification task for ResNet-50 with 91.60% accuracy and CutMix yields the greatest improvement for ViT-B with 97.94% accuracy. Code will be available at https://github.com/AIGeeksGroup/MediAug.

MediAug: Exploring Visual Augmentation in Medical Imaging

TL;DR

Medical image augmentation faces a domain gap with natural images and fragmented evaluation across tasks. MediAug provides a unified benchmark applying six mix-based augmentations—MixUp, YOCO, CropMix, CutMix, AugMix, and SnapMix—across ResNet-50 and ViT-B backbones on brain-tumor MRI and eye-fundus datasets, enabling direct cross-method comparisons. Results show task-specific best pairings (MixUp with ResNet-50 and SnapMix with ViT-B for brain tumors; YOCO with ResNet-50 and CutMix with ViT-B for eye diseases) and underscore the importance of hyperparameter tuning (e.g., CutMix ). The framework, with public code, offers practical, reproducible guidance for robust, generalizable clinical AI systems and accelerates the safe deployment of deep learning in medical imaging.

Abstract

Data augmentation is essential in medical imaging for improving classification accuracy, lesion detection, and organ segmentation under limited data conditions. However, two significant challenges remain. First, a pronounced domain gap between natural photographs and medical images can distort critical disease features. Second, augmentation studies in medical imaging are fragmented and limited to single tasks or architectures, leaving the benefits of advanced mix-based strategies unclear. To address these challenges, we propose a unified evaluation framework with six mix-based augmentation methods integrated with both convolutional and transformer backbones on brain tumour MRI and eye disease fundus datasets. Our contributions are threefold. (1) We introduce MediAug, a comprehensive and reproducible benchmark for advanced data augmentation in medical imaging. (2) We systematically evaluate MixUp, YOCO, CropMix, CutMix, AugMix, and SnapMix with ResNet-50 and ViT-B backbones. (3) We demonstrate through extensive experiments that MixUp yields the greatest improvement on the brain tumor classification task for ResNet-50 with 79.19% accuracy and SnapMix yields the greatest improvement for ViT-B with 99.44% accuracy, and that YOCO yields the greatest improvement on the eye disease classification task for ResNet-50 with 91.60% accuracy and CutMix yields the greatest improvement for ViT-B with 97.94% accuracy. Code will be available at https://github.com/AIGeeksGroup/MediAug.
Paper Structure (20 sections, 17 equations, 92 figures, 4 tables)

This paper contains 20 sections, 17 equations, 92 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Overview
  5. MixUp
  6. YOCO
  7. CropMix
  8. CutMix
  9. AugMix
  10. SnapMix
  11. Experiments
  12. Datasets and Evaluation Matrices
  13. Brain Tumor Classification Dataset
  14. Eye Diseases Classification Dataset
  15. Evaluation Matrices:
  16. ...and 5 more sections

Figures (92)

  • Figure 1: Architecture of MediAug: We enhance medical representation learning via advanced visual augmentation.
  • Figure 1: Flippingis a data augmentation technique that generates a mirror image by inverting pixel positions across a specified axis, typically horizontally, to expand training datasets and improve model generalization.
  • Figure 2: Augmentation in Eye diseases classification dataseteye_diseases_classification_dataset: (a) An image of a cataract. (b) MixUpzhang2017mixup: Original image and contrast-enhanced image is mixed with a mixing parameter $\lambda = 0.42$. (c) YOCOhan2022you: splits, flips, and recombines the image for augmentation. (d) CropMixhan2022cropmix: combines three 25% cropped views using MixUp zhang2017mixup for augmentation. (e) CutMixyun2019cutmix: augments by relocating a 1/4 cropped region within the image. (f) AugMixhendrycks2019augmix: AugMix blends blurring, sharpening, and color adjustments with the original. (g) SnapMixhuang2021snapmix: SnapMix blends interpolated regions, weighted by saliency maps, with the original.
  • Figure 2: Rotation based image augmentation transforms the original image by pivoting it around its centre across an axis within 1-359 degrees while preserving pixel relationships.
  • Figure 3: Augmentation in Brain Tumor Classification (MRI) datasetsartaj_bhuvaji_ankita_kadam_prajakta_bhumkar_sameer_dedge_swati_kanchan_2020: (a) An image of a no_tumor in Testing. (b) MixUpzhang2017mixup: Contrast-adjusted and mixed ($\lambda=0.30$), in grayscale and pseudo-color. (c) YOCOhan2022you: The image was resized, contrast-enhanced, flipped, and concatenated. (d) CropMixhan2022cropmix: It from a contrast-enhanced, randomly cropped image using CutMix. (e) CutMixyun2019cutmix: Original image flipped and mixed using random CutMix enhancement ($\lambda=0.70$). (f) AugMixhendrycks2019augmix: It enhances the image with flipping and brightness adjustment. (g) SnapMixhuang2021snapmix: It enhances by blending resized cropped regions with the original image.
  • ...and 87 more figures