Hyperspectral angular domain imaging for ex-vivo breast tumor detection Fartash Vasefi a, b, c , Bozena Kaminska c , Muriel Brackstone d, e , and Jeffrey J. L. Carson a, b,* a Department of Medical Biophysics, Schulich School of Medicine and Dentistry, University of Western Ontario, London, ON, Canada b Imaging Program, Lawson Health Research Institute, London, ON, Canada c The School of Engineering Science, Simon Fraser University, Burnaby, BC, Canada d London Regional Cancer Program, London, Ontario, Canada e Department of Surgery, University of Western Ontario, London, Ontario, Canada (Corresponding author: jcarson@lawsonimaging.ca) ABSTRACT An angular domain spectroscopic imaging system was built and validated using fresh ex vivo breast tissue samples (~ 2 mm thick). The hyperspectral system consisted of a halogen lamp, collimation optics, scanning stage with controller, a silicon micro-machined micro-channel array, and a pushbroom spectral imager. As a proof of concept, spectral data cubes acquired from tissue samples were input into principal component analysis and Mahalanobis discriminant analysis to differentiate between spectral signatures of breast tumor and normal tissue. It is proposed that the results from training sets can be used to construct a set of classifiers to enable tumor detection in samples representative of the surgical margins. Keywords: Breast cancer; trans-illumination imaging; hyperspectral imaging, principal component analysis 1. INTRODUCTION Breast conserving surgery (i.e. lumpectomy) followed by radiation is considered the recommended surgical approach over mastectomy for women diagnosed with early breast cancer. A recent study of 112,154 women diagnosed with stage I or II breast cancer compared lumpectomy/radiation (55%) versus mastectomy (45%) procedures and found that the survival rate for women who received lumpectomy/radiation was more favorable regardless of age or cancer type [1]. Successful breast conserving surgery is directly dependent on complete removal of the tumor mass with clear margins [2]. Should tumor tissue be discovered at or nearby the margins, then a second (or sometimes third) surgery is required to clear the margins of tumor tissue. Recent studies have shown that typically 20-40% of patients undergoing lumpectomy require second surgeries because of a close or positive surgical margin diagnosed post-operatively [3]. The current gold standard for differentiating between normal and tumor tissue during surgery is by histopathology assessment that includes high-resolution microscopy of thinly sliced and stained tissue. This process is time-consuming with a high chance of under-sampling. Many areas are left microscopically uninspected since the microscopic field of view is limited. Therefore, the intra-operative microscopy approach is not ideal for testing the entire three-dimensional volume which would be more desirable [4] as well as effective [5]. Alternatively, ex vivo specimen x-ray based imaging systems used at the point-of-care (e.g. Faxitron BioVision imaging system, Faxitron Bioptics, Tucson, AZ) may substantially reduce re-excision rates [8]. However, it is still difficult to identify the tumor margins efficiently in breast carcinomas with non-grossly recognizable abnormalities such as microcalcifications using x-ray based systems [9]. Also, comparison of x-ray procedures to frozen section analysis suggest that it is still not clear whether the x-ray radiography system provides better margin detection accuracy during breast conservation therapy procedures [10]. Many researchers have studied optical imaging due to its excellent tissue contrast compared to other imaging modalities. In the last decade, methods have been developed for intra-operative tumor margin detection utilizing diffuse reflectance spectroscopy [11], intrinsic fluorescence spectroscopy [12], optical coherence tomography [13], Terahertz pulsed spectroscopy [14], and Raman spectroscopy [15]. It is known that a number of cancer types exhibit different spectral signatures compared to the surrounding normal tissue [16][17][18]. Hyperspectral imaging with subsequent Mahalanobis discriminant analysis has been applied to ex vivo unstained breast tumor samples from a small animal model [19]. Imaging, Manipulation, and Analysis of Biomolecules, Cells, and Tissues XI, edited by Daniel L. Farkas, Dan V. Nicolau, Robert C. Leif, Proc. of SPIE Vol. 8587, 85870S © 2013 SPIE · CCC code: 1605-7422/13/$18 · doi: 10.1117/12.2004663 Proc. of SPIE Vol. 8587 85870S-1 Downloaded From: http://spiedigitallibrary.org/ on 05/14/2013 Terms of Use: http://spiedl.org/terms