Treating the Image as a System: A Hybrid Signal-AI Approach to Mammogram Analysis

Early detection of breast cancer is crucial, yet standard mammography faces challenges with diagnostic accuracy. While deep learning models offer high performance, their ‘black box’ nature often hinders clinical trust and interpretability. This study proposes a novel hybrid diagnostic model that integrates signal processing and deep learning under a systems theory framework. We treat mammograms as complex 2D signals and extract interpretable features using Radon transforms, Gabor filters, and Fast Fourier Transform (FFT). These handcrafted features are then fused with high-level abstract features derived from a fine-tuned ResNet50 architecture. The proposed hybrid model, evaluated on a public mammography dataset of 24,576 images, demonstrated state-of-the-art performance. It achieved an overall accuracy of $99.74 \%$, with a malignant class recall of $99.78 \%$ and an Area Under the Curve (AUC) of 1.00, significantly outperforming baseline models. Our findings demonstrate that this synergistic fusion, guided by systems theory, creates a diagnostic tool that is not only highly accurate but also enhances interpretability, presenting a promising pathway toward developing more reliable and clinically trusted AI-powered decision support systems for breast cancer diagnosis.