Automated Breast Cancer Classification Using Optical Tomographic Images

In this paper, an automated procedure is implemented for detecting breast cancers based on optical tomographic images obtained by a phase-contrast diffuse optical tomography (PCDOT) system, which can measure the refractive indices of human breast masses in vivo, in addition to the absorption and scattering coefficients obtained by a traditional diffuse optical tomography (DOT) system. This classification procedure automatically extracts absorption, scattering, and refractive index attributes from optical tomographic images and applies a support vector machine (SVM) classifier to distinguish the malignant images from the benign ones based on these automatically extracted attributes. The sensitivity, specificity, and overall accuracy of the classification results using absorption, scattering, and refractive index attributes extracted from the optical tomographic images of 35 human breast masses are 81.8%, 91.7%, and 88.6%, respectively. Conversely, the sensitivity, specificity, and overall accuracy of the classification results using only the absorption and scattering attributes are 63.6%, 83.3%, and 77.1%, respectively. These results indicate that the PCDOT system, which can measure the refractive index in addition to absorption and scattering coefficients, combined with automated classification procedure achieves better performance in detecting breast cancers than the traditional DOT system does. Furthermore, the results obtained by our automated classification procedure are also better than the classification results obtained by an experienced technician through visual examination. The sensitivity, specificity, and overall accuracy of the classification based on the visual examination are 81.8%, 70.8%, and 74.3% respectively.