GEMNet: Gabor Enhanced Multiscale CNN for Breast Tumor Detection Using Biorthogonal Wavelet Transform Features

Breast cancer remains a prevalent cause of mortality, and accurate early detection through mammography significantly improves patient outcomes. However, current diagnostic techniques often struggle with distinguishing benign from malignant mammographic findings, particularly in dense or subtle presentations. To address these limitations, we propose a novel deep learning framework, the Gabor Enhanced Multi-Scale Neural Network (GEMNet), integrated with biorthogonal wavelet transforms to capture fine-grained spatial and frequency domain features. The wavelet transforms enable multi-resolution analysis, allowing the model to identify localized variations in texture and intensity, which are critical for accurate lesion classification. Our approach leverages Gabor filters for enhanced feature extraction, enabling the model to discern subtle patterns within mammograms. We rigorously benchmarked GEMNet against leading architectures, including EfficientNet B0, ResNet150, VGG16, and MobileNetV2, across comprehensive mammographic datasets. Results indicate that GEMNet outperforms these state-of-the-art models, achieving an F1 score of 97% and an AUC of 99.3%, marking a significant advancement in diagnostic accuracy. This work demonstrates the efficacy of combining multi-scale analysis with wavelet-based transformations, providing a robust framework for high-precision mammographic lesion classification. GEMNet holds promise for integration into clinical workflows, potentially reducing false positives and false negatives, and advancing breast cancer diagnostics through machine learning.