Polarization imaging for breast cancer diagnosis using texture analysis and SVM

The polarization state of the light transmitting through a specimen can provide useful information concerning the optical property of the sample. We present a polarization imaging device that measures the Stokes components of the trans-illuminant light. Particularly, we develop an image analysis algorithm using texture analysis and support vector machine for tissue classification. The experiment has been conducted using rat breast tissue in health and with cancer. It is demonstrated that the Stokes images can improve the classification performance of the algorithm. With the incorporation of the multi-polarization images, it is shown that the diagnosis accuracy can be further improved over that using intensity image only.