<p>Breast cancer detection is a major global health challenge since traditional imaging methods, such as mammography and ultrasound, have certain limitations, including ionizing radiation, operator dependency, limited contrast, and high operational cost. TeraHertz (THz) electromagnetic sensing offers a promising alternative because of its non-ionizing nature and high sensitivity to dielectric variations in biological tissues. In the literature, most THz research focuses on antenna design rather than automated diagnostic frameworks that analyse complex spectral patterns for accurate tumour detection and classification. To address this gap, the research introduces THzSenseNet, a lightweight THz wrench-shaped patch antenna-assisted machine learning model designed to classify various breast tissues, including healthy, benign, and malignant, using scattering-parameter signatures. THz multiple-input and multiple-output antenna integrates multilayer dual-resonant, structured spectral preprocessing and a hybrid multi-scale spectral convolutional encoder/bidirectional spectral memory unit classifier optimized through the revolution optimization algorithm. Using simulation-derived datasets representing diverse tumor sizes and depths, THzSenseNet achieves 99% accuracy, surpassing existing methods by 2.7%. Results confirm that combining multi-resonant THz sensing with advanced spectral-temporal learning offers a highly sensitive, non-ionizing, and computationally efficient pathway for early breast cancer detection. Outcomes underscore the THzSenseNet potential as a strong foundation for future clinical validation and portable THz diagnostic technologies.</p>