SENet-Augmented Explainable Deep Feature Framework with Machine Learning for Breast Tumor Detection in Ultrasound Imaging
Mohammed Ibrahim Hussain, Safiul Haque Chowdhury, Md. Shovon, Monoara Sultana Morzina, Muhammad Minoar Hossain, Mohammad Mamun · 2025
Breast tumors pose a significant threat to women's health globally, with increasing diagnosis rates underscoring the need for early and accurate detection. Early identification of tumors plays a critical role in reducing risks associated with non-cancerous and cancerous growths. This study proposes an interpretable and robust framework for multi-class breast tumor classification using ultrasound images. The dataset comprises 1,578 breast ultrasound images categorized into malignant, benign, and normal. We preprocess all images by resizing them to 256×256 pixels and applying Speckle Reducing Anisotropic Diffusion (SRAD) filtering to enhance image quality. Data augmentation techniques are employed to increase variability and improve model generalization. We extract deep features using four Convolutional Neural Network (CNN) architectures: VGG-16, VGG-19, ResNet-50, and ResNet-101. We integrate a Squeeze-and-Excitation Network (SENet) block with each deep learning (DL) model to enhance feature representation. The extracted features are then passed to five traditional Machine Learning (ML) classifiers: Random Forest (RF), Extreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGBM), Gradient Boosting Decision Trees (GBDT), and Categorical Boosting (CB), to perform classification into normal, benign, or malignant categories. To ensure model transparency and trustworthiness, we apply state-of-the-art Explainable Artificial Intelligence (XAI) techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), Grad-CAM++, and Local Interpretable Model-agnostic Explanations (LIME), to visualize and interpret the model's decision-making process. Among all combinations, the ResNet-50 with XGB achieves the highest classification accuracy of 99.76%, demonstrating the effectiveness of our hybrid approach in accurately and interpretably detecting breast tumors from ultrasound images.