An artificial intelligence model for early-stage breast cancer classification from histopathological biopsy images
Neil Chaudhary, A. Z. Dhunny
Abstract
Accurate identification of breast cancer subtypes is essential for guiding treatment decisions and improving patient outcomes. In current clinical practice, determining histological subtypes often requires additional invasive procedures, delaying treatment initiation. This study proposes a deep learning-based model built on a DenseNet121 backbone with a multi-scale feature fusion strategy, designed to classify breast cancer from histopathological biopsy images. Trained and evaluated on the publicly available BreaKHis dataset using 5-fold cross-validation, the model achieved a binary classification accuracy of 97.1%, and subtype classification accuracies of 93.8% for benign tumors and 92.0% for malignant tumors. These results demonstrate the model's ability to capture morphological cues at multiple levels of abstraction and highlight its potential as a diagnostic support tool in digital pathology workflows.