Performance benchmarking of convolutional neural networks and ensemble machine learning techniques for automated mammographic breast cancer detection: A comparative study

Breast cancer remains a leading cause of mortality among women worldwide, making early and accurate detection crucial for improving patient outcomes. This study presents a comparative analysis of various machine learning algorithms—EfficientNet, DenseNet, ResNeXt-50, and Support Vector Machine (SVM)—in predicting breast cancer using mammography images. Utilizing a dataset of 5,339 mammograms from the Digital Database for Screening Mammography (DDSM), the models were trained and tested on two classes: benign and malignant lesions. The mammograms underwent preprocessing techniques, including image quality assessment, contrast enhancement, and artifact removal, to ensure high-quality data for model training. The performance of each model was evaluated using metrics such as accuracy, sensitivity, specificity, and ROC analysis. The results revealed that the EfficientNet model outperformed the other algorithms, achieving an accuracy of 95.23%, sensitivity of 96.67%, and specificity of 93.82%. In contrast, DenseNet exhibited the lowest performance, struggling with the correct classification of cancer cases. The comparative analysis highlights the strengths and weaknesses of each model, offering valuable insights into their potential clinical applications. This research underscores the importance of selecting the appropriate machine learning architecture to enhance the predictive accuracy of breast cancer detection and provides a foundation for integrating these models into clinical practice for personalized treatment planning. Future studies will focus on expanding the dataset and improving model generalizability to diverse patient populations.