Optical spectral surveillance of breast tissue landscapes for detection of residual disease in breast tumor margins.

Abstract

We demonstrate a strategy to "sense" the micro-morphology of a breast tumor margin over a wide field of view by creating quantitative hyperspectral maps of the tissue optical properties (absorption and scattering), where each voxel can be deconstructed to provide information on the underlying histology. Information about the underlying tissue histology is encoded in the quantitative spectral information (in the visible wavelength range), and residual carcinoma is detected as a shift in the histological landscape to one with less fat and higher glandular content. To demonstrate this strategy, fully intact, fresh lumpectomy specimens (n = 88) from 70 patients were imaged intra-operatively. The ability of spectral imaging to sense changes in histology over large imaging areas was determined using inter-patient mammographic breast density (MBD) variation in cancer-free tissues as a model system. We discovered that increased MBD was associated with higher baseline β-carotene concentrations (p = 0.066) and higher scattering coefficients (p = 0.007) as measured by spectral imaging, and a trend toward decreased adipocyte size and increased adipocyte density as measured by histological examination in BMI-matched patients. The ability of spectral imaging to detect cancer intra-operatively was demonstrated when MBD-specific breast characteristics were considered. Specifically, the ratio of β-carotene concentration to the light scattering coefficient can report on the relative amount of fat to glandular density at the tissue surface to determine positive margin status, when baseline differences in these parameters between patients with low and high MBD are taken into account by the appropriate selection of threshold values. When MBD was included as a variable a priori, the device was estimated to have a sensitivity of 74% and a specificity of 86% in detecting close or positive margins, regardless of tumor type. Superior performance was demonstrated in high MBD tissue, a population that typically has a higher percentage of involved margins.