Abstract Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide. Early and accurate detection is critical for improving survival rates and enabling effective treatment. This research proposes a fine-tuned Multi-Deep Neural Network model with Multiple Datasets (FT-MDNNMDs). The proposed system employs a hierarchical classification strategy, first distinguishing cancer from normal cases, and subsequently classifying cancer-positive cases into Stage II and Stage III categories. The model integrates data from The Cancer Genome Atlas (TCGA) and a private clinical image dataset collected from hospitals in Pakistan. Preprocessing techniques, including image normalization, filtering, and Principal Component Analysis (PCA), are applied to enhance feature quality and reduce redundancy. Transfer learning and fine-tuning strategies are incorporated to further improve classification performance. The proposed model effectively distinguishes between benign and malignant tumors and accurately identifies stage II and stage III cases. Experimental results demonstrate that the fine-tuned MDNNMDs model achieves an accuracy of 99.57%, outperforming conventional machine learning algorithms such as Support Vector Machine (94.46%), Decision Tree (93.54%), and Naïve Bayes (91.22%). The model also achieved superior MCC (99.46%), F-score (99.57%), and recall (99.63%). In addition to algorithmic comparison, the performance of the proposed model is compared with existing breast cancer detection models, and it is revealed that the proposed model provides more accurate results than existing ones, with 1.5% to 3.0% better accuracy.
Shafique et al. (Wed,) studied this question.