A Hybrid Two-Step Approach For Breast Cancer Classification In Low-Resource Settings
Gergő Bogacsovics · 2025
We present a hybrid two-step approach that combines deep learning- and machine learning-based approaches for the accurate and reliable classification of breast cancer using digitized ultrasound images in a low-resource setting. The proposed method first trains multiple convolutional neural networks (CNNs) on the dataset and then uses their convolutional layers to extract the most important features and patterns in the given input image. Then, these features are merged and passed to a support vector machine (SVM) model, which turns them into the final output of the model. During our experiments, we considered multiple state-of-the-art CNN architectures. We show that our proposed model can overcome the biggest disadvantage of deep learning-based models which usually underperform if the size of the dataset is small. Namely, we showcase that our hybrid model can detect breast cancer with high accuracy and sensitivity, surpassing all of the CNN architectures and other state-of-the-art methods as well, achieving 90% accuracy, 88.9% precision, 89.3% sensitivity, and 88.7% F1-score.