Enhancing IDC Histopathology Image Classification: A Comparative Study of Fine-Tuned and Pre-trained Models

Invasive ductal carcinoma (IDC) is a type of breast cancer that affects adult women all around the world. This cancer starts in the duct cells of the breast, spreads through the lymph system, and eventually affects nearby organs and bones. It’s crucial for physicians to correctly identify the various forms of breast cancer. Instead of doing this manually, it’s better to use computer programs because it saves time and reduces mistakes. This study presents a computer-assisted diagnosis method that utilizes deep convolutional neural networks to classify IDC histopathology images. These networks are trained using two forms of transfer learning: feature extraction and fine-tuning. In this study IDC classification is done using 4 well known deep learning networks, Xception, DenseNet169, ResNet101 and MobileNetV2. The dataset used is a publicly available IDC dataset containing 168 whole slide images. The evaluation results show that the fine-tuned models give better classification results than feature extractor models for IDC histopathology image classification.