Temporal Dynamics in Microwave Breast Imaging: Enhancing Cancer Detection with Transformers

Microwave breast imaging (MWBI) constitutes a well-promising alternative to conventional imaging modalities such as mammography and ultrasound, which could offer a non-invasive, low-cost, radiation-free screening modality. This work aims to determine the capability of exploiting temporal dynamics in MWBI data by state-of-the-art deep learning schemes, especially transformer architectures. We present a comparative study of several ML and DL techniques, including the Random Forest, Support Vector Machine, Logistic Regression, ۱D CNNs, and transformers, to detect breast cancer. The results showed that transformer models would depict complicated temporal dependencies with the highest accuracy of ۹۱.۲۰% and sensitivity of ۹۸.۸۰%, while outperforming all the other methods. It also evidences that temporal information plays a crucial role in MWBI signals and positions transformers as a strong solution for the early diagnosis of breast cancer. The study will open up possibilities for advanced DL techniques within an MWBI system, especially in resource-constrained settings.

کلیدواژه ها:

Microwave Breast Imaging ، Deep Learning ، Machine Learning ، Convolutional Neural Networks ، Transformers