DieleNet: Dielectric properties based breast cancer classification using deep learning

Irregular cells grow out of control in the breast, leading to breast cancer, which is one of the primary cancers affecting women's health. This work uses our real-time dielectric signature measurements of breast cell lines collected from the Open-Ended Coaxial Probe (OECP) to detect breast cancer effectively. The dataset comprises five breast cell lines – four cancerous and one normal – recorded with and without a culture medium. Existing approaches from the literature often struggle with manual feature extraction, poor handling of sequential data, and limited effectiveness when applied to real-time databases. To overcome these problems, we proposed a Dielectric-based Breast Cancer Classification Network (DieleNet) for abnormality detection. A deep learning architecture is proposed with an attention mechanism as a backbone for the convolution and long-short-term memory combination, where we can extract frequency-specific dielectric variations and complex dispersion patterns across broad frequency bands. Additionally, the attention mechanism dynamically weights the contribution of each time step (frequency point), allowing the model to focus on the most informative spectral regions relevant to classification. This makes it highly suitable for modeling breast cancer cell lines’ complex, nonlinear dielectric behavior, measured using the OECP technique. The proposed model achieved an accuracy of 97.81%±0.24 with media and 97.88%±0.91 without media in the classification of breast cancer cells.