Class-Wise Combination of Mixture-Based Data Augmentation for Class Imbalance Learning of Focal Liver Lesions in Abdominal CT Images

• DOI: 10.1007/s10278-025-01415-8
• Publication Date: 2025-01-27T19:29:56Z
• Authors: Hansang Lee; Deokseon Kim; Joonseok Lim; Helen Hong

In this paper, we propose a method to address the class imbalance learning in classification of focal liver lesions (FLLs) from abdominal CT images. Class is significant challenge medical image analysis, making it difficult for machine models learn classify them accurately. To overcome this, class-wise combination mixture-based data augmentation (CCDA) that uses two techniques, MixUp and AugMix. These are applied at different ratios each adaptively features class. This tailored handle unique characteristics FLL type by adjusting mix major classes (e.g., cysts metastases) minor hemangiomas). experiments, our was validated on dataset consisted portal phase images 1290 colorectal cancer patients. The results showed applying AugMix manner could significantly improve performance while maintaining or slightly improving classes. Quantitative higher F1 scores more balanced accuracy across when CCDA used compared other methods.