Breast conserving surgery (BCS) is a recommended treatment for early stage breast cancer and for breast cancers that have been reduced in size by neoadjuvant therapy. In BCS (also known as a partial mastectomy or lumpectomy), the surgeon attempts to excise the tumor along with a margin of normal tissue, while preserving as much of the normal breast tissue as possible. Approximately 160,000 patients are eligible for breast conserving therapy each year and as many as 20-70% of patients undergoing BCS require repeat surgeries due to a close or positive surgical margin diagnosed post-operatively [1
]. The pathologic margin status is an important predictor of local recurrence of an invasive or in situ
cancer after BCS [5
]. Thus, complete excision of the tumor is essential to reduce the risk of recurrence [7
Currently, surgeons do not have adequate intra-operative assessment tools to ensure that the cancer has been completely removed at the time of first surgery. The lack of this capability represents a significant unmet clinical need. Only a small number of hospitals who perform BCS (less than 5%, including the Moffitt Cancer Center in Tampa, FL and the MD Anderson Cancer Center in Houston, TX) currently utilize intra-operative cytologic or pathologic analysis of tumor margins. Touch-prep cytology is a technique in which cells on the surface of the tissue are transferred to glass slides by touching the specimen to the glass, and are then stained for pathologic observation. Touch-prep cytology allows for evaluation of the whole lumpectomy surface, albeit with a wide range of sensitivities (38-100%) and specificities (85-100%) reported in the literature [8
]. Furthermore, this technique is time consuming, requires special expertise by a cytopathologist, and does not detect tumor cells close to the lumpectomy surface. Frozen section analysis, in which the tissue is frozen and select microscopically thin sections are cut from the specimen for pathologic observation, is a technically challenging procedure due to the significant amount of fatty tissue found in breast specimens. Sensitivity ranges in the literature from 65 to 91% and specificity ranges from 86 to 100% [11
A fast, non-destructive device that could image breast tumor margins in the operating room would be highly desirable to ensure complete removal of the cancer and thus reduce the risk of local recurrence. The device needs to 1) be capable of surveying multiple margins in an acceptable amount of time (within 20 minutes which is the amount of time it takes frozen section) [14
], 2) have a sensing depth of 0-2 mm (the accepted criterion for clear margins) [15
], 3) cover a large area (the majority of margin areas range from ~10-20 cm2
in our study), 4) image with a resolution on the order of millimeters (comparable to the thickness of bread loafed slices evaluated by pathology), and 5) effectively detect differences between benign and malignant tissues and to do this without the need for pathologic evaluation, or tissue processing.
Optical imaging of tumor margins is attractive because it can quickly sample an entire tumor margin intra-operatively without damaging the tissue. Several groups are working on optical techniques for breast tumor margin assessment. Bigio et al used reflectance spectroscopy in a preliminary study to look at in vivo
sites on the tumor bed in 24 patients (13 cancer and 59 normal sites). They showed that using hierarchical cluster analysis, cancer and normal sites could be separated with a sensitivity of 67% and a specificity of 79% [20
]. Haka et al recently published on Raman spectroscopy to prospectively examine freshly excised lumpectomy specimens, which were sliced to expose tumor sites in 21 patients (123 benign and 6 malignant tissue sites) and reported a sensitivity of 83% and a specificity of 93% [21
]. Their previous retrospective study showed 94% sensitivity and 96% specificity for ex viv
o measurements of frozen samples [22
]. Volynskaya et al demonstrated the ability of diffuse reflectance spectroscopy and intrinsic fluorescence spectroscopy to differentiate various benign and malignant tissues in breast biopsies from 17 patients (95 benign and 9 malignant sites), resulting in a sensitivity of 100% and a specificity of 96% [23
]. Nguyen et al demonstrated that optical coherence tomography can detect tumor margin positivity in 20 patients (9 positive and 11 negative margins) with a sensitivity and specificity of 100% and 82%, respectively [24
]. Keller et al recently published work on diffuse reflectance and fluorescence spectroscopy to detect cancerous sites on breast tumor margins in 32 patients (145 normal and 34 individual tumor sites), and reported a sensitivity and specificity of 85% and 96%, respectively, for classifying individual sites [25
Our group has developed a first-generation optical spectral imaging platform that operates in the visible spectral range (450-600 nm) to rapidly and non-destructively create molecular composition maps of the tumor margin. The technologies presented above are restricted to sampling a very small area of the margin and therefore do not have the capability to image full tumor margins which is critical for a margin assessment device. We believe that this is a differentiating feature of our technology. The optical spectral imaging platform is based on diffuse reflectance spectroscopy. Diffuse reflectance spectroscopy measures the remitted light as a function of wavelength and the magnitude and shape of the spectrum is reflective of the absorption and scattering properties of the tissue. Our group has also developed a fast, scalable Monte Carlo model [26
] to reliably and quantitatively determine the wavelength dependent absorption and reduced scattering coefficients of the tissue (µa
’ respectively) from the diffuse reflectance spectra measured with the optical spectral imaging system. The concentration of the absorbers can be easily derived from the absorption coefficient spectra using the Beer-Lambert equation. The primary absorbers in the breast over the visible spectral range are oxygenated hemoglobin, deoxygenated hemoglobin, and β-carotene. The primary scatterers reflected by the scattering coefficient are cells and sub-cellular organelles. These extracted parameters can be used to create maps of tissue composition of the breast tumor margins.
The optical spectral imaging device was used in a clinical study from December 2007 to June 2009 on 120 patients undergoing BCS. The purpose of the study was to determine the feasibility of the device for the detection of close/positive tumor margins in an intra-operative setting. In this study, optical spectral images were collected from 1 to 2 margins per patient. The four corners of each imaged margin were inked such that the extracted parameter maps for each margin could be compared to the overall diagnosis of that margin, based on routine margin-level pathology (this is referred to as margin-level analysis). At Duke University Medical Center (DUMC), a margin is considered positive if there are tumor cells touching ink, close if there are tumor cells within 2 mm of the ink, and negative (or clear) if tumor cells are > 2mm. In addition, 6-10 sites within each imaged margin were also randomly inked such that the extracted parameters collected from those 6-10 pixels could be directly compared with pathology (this is referred to as site-level pathology).
Wilke et al [2
] reported on an initial subset of patients (n = 48) from the above study where the extracted parameter maps were used as the basis for a classification scheme to detect margin positivity. The gold standard in this case was margin-level pathology. The classification scheme based on the extracted parameter maps accurately identified 79.4% of the pathologically close/positive margins and had a specificity of 66.7%. These close/positive margins included several types of malignancies, mainly ductal carcinoma in situ
and invasive ductal carcinoma but also lobular cancer, lobular carcinoma in situ
, and tubular cancer. This initial study showed that the sensitivity of the technology is comparable to currently available intra-operative margin assessment tools such as frozen section but has the benefits of not requiring any type of tissue cutting, preparation, or a pathologist in the operating room.
The goal of this paper was to establish the performance metrics of the optical spectral imaging system we have developed in a manner that is relevant to breast tumor margin assessment. Specifically, this paper quantifies important sources of systematic and random errors that could arise when the system is used in a clinical setting. The endpoints characterized in this paper are, the SNR of the system, the accuracy with which the device characterizes the composition of tumor margins, the sensing depth, the amount of crosstalk between adjacent channels of the probe, and reproducibility. The optical properties of histologically normal and malignant breast tissues obtained from the randomly inked sites (site-level analysis) served as the basis for characterizing the instrument performance metrics enumerated above. Since the clinical aspects of the study are already described by Wilke et al [2
], this paper will focus more on the technological aspects of the intra-operative margin assessment project. The methods section is organized by a description of the instrumentation, as well as the clinical and pathological procedures. Next, the data analysis, experiments and simulations specifically addressed in this paper are described.