Details of Research Outputs

Breast cancer has become the leading cause of women cancer death worldwide. Despite the consensus that breast cancer early detection can significantly reduce treatment difficulty and cancer mortality, people still are reluctant to go to hospital for regular checkups due to the high costs incurred. A timely, private, affordable, and effective household breast cancer early detection solution is badly needed. In this paper, we propose a household solution that utilizes pairs of sensors embedded in the bra to measure the thermal and moisture time series data (BTMTSD) of the breast surface and conduct time series classification (TSC) to diagnose breast cancer. Three main challenges are encountered when doing BTMTSD classification, (1) small supervised dataset, which is a common limitation of medical research, (2) noisy time series with unique noise patterns, and (3) complex interplay patterns across multiple time series dimensions. To mitigate these problems, we incorporate multiple data augmentation and transformation techniques with various deep learning TSC approaches and compare their performances for the BTMTSD classification task. Experimental results validate the effectiveness of our framework in providing reliable breast cancer early detection.