上海科技大学知识管理系统

Objectivee：Breast cancer poses a significant threat to women's health. Microwave-induced thermoacoustic imaging (MITAI) technology can be utilized for detecting tumors within the breast. However, due to the limitations of the irradiation area of traditional single antenna systems, uneven microwave energy distribution often occurs, making it difficult to detect tumors in certain regions. The aim of this study is to propose a microwave-focused thermoacoustic imaging (MF-TAI) method to address the issue of uneven microwave energy distribution and to employ deep learning techniques for tumor detection. Methods ：We employ advanced phased array antenna technology to achieve highly focused microwave energy distribution within various target regions, enabling imaging through Microwave-Induced Thermoacoustic Tomography (MITAT). To enhance the process, we introduce a transfer learning approach, leveraging pre-trained model weights to reduce training time and resource consumption. The model is trained using conventional thermoacoustic images as input, which are then compared to simulated true tumor labels to learn the mapping relationship, thereby achieving high-precision detection. Results ：Using both simulated and experimentally obtained images as input data for the deep learning model, we validate its tumor detection capabilities. The results demonstrate that the fully trained Res-Net model is capable of accurately detecting the presence or absence of tumors in both simulated and experimental images. Conclusion ：By integrating MF-TAI technology with transfer learning, this study resolves uneven microwave irradiation in traditional MITAI, enabling precise breast tumor detection. This technical framework provides powerful technical support for early diagnosis and treatment of breast cancer.