Breast Tumor Classification Using EfficientNet Deep Learning Model

TL;DR

This work adopted EfficientNet, a state-of-the-art convolutional neural network model that balances high accuracy with computational cost efficiency, and introduced an intensive data augmentation pipeline and cost-sensitive learning, improving representation and ensuring that the model does not overly favor majority classes.

Abstract

Precise breast cancer classification on histopathological images has the potential to greatly improve the diagnosis and patient outcome in oncology. The data imbalance problem largely stems from the inherent imbalance within medical image datasets, where certain tumor subtypes may appear much less frequently. This constitutes a considerable limitation in biased model predictions that can overlook critical but rare classes. In this work, we adopted EfficientNet, a state-of-the-art convolutional neural network (CNN) model that balances high accuracy with computational cost efficiency. To address data imbalance, we introduce an intensive data augmentation pipeline and cost-sensitive learning, improving representation and ensuring that the model does not overly favor majority classes. This approach provides the ability to learn effectively from rare tumor types, improving its robustness. Additionally, we fine-tuned the model using transfer learning, where weights in the beginning trained on a binary classification task were adopted to multi-class classification, improving the capability to detect complex patterns within the BreakHis dataset. Our results underscore significant improvements in the binary classification performance, achieving an exceptional recall increase for benign cases from 0.92 to 0.95, alongside an accuracy enhancement from 97.35 % to 98.23%. Our approach improved the performance of multi-class tasks from 91.27% with regular augmentation to 94.54% with intensive augmentation, reaching 95.04% with transfer learning. This framework demonstrated substantial gains in precision in the minority classes, such as Mucinous carcinoma and Papillary carcinoma, while maintaining high recall consistently across these critical subtypes, as further confirmed by confusion matrix analysis.

Figures (7)

• Figure 1: Schematic of the proposed system workflow. Section A involves data preprocessing, including splitting the dataset into training and testing sets, followed by two levels of augmentation: intensive data augmentation for underrepresented classes and normal augmentation for the remaining data. Section B highlights the model architecture, utilizing EfficientNet with cost-sensitive learning. Section C covers the final layers of the model.
• Figure 2: Validation accuracy across three augmentation parameter intensities.
• Figure 3: Augmented histopathological breast tissue images. The first row displays Adenosis and Phyllodes Tumor images, and the second row shows Tubular Adenoma images, each subjected to intensive augmentation techniques, including horizontal and vertical flips, affine transformations, brightness adjustment, Gaussian blur, and additive Gaussian noise.
• Figure 4: Confusion matrices for binary classification results. EfficientNet B5 (a) without and (b) with intensive augmentation.
• Figure 5: Confusion matrix illustrating the results of multi-class classification with normal data augmentation.