Raman Spectroscopy for Rapid Evaluation of Surgical Margins during Breast Cancer Lumpectomy

Abstract

Failure to precisely distinguish malignant from healthy tissue has severe implications for breast cancer surgical outcomes. Clinical prognoses depend on precisely distinguishing healthy from malignant tissue during surgery. Laser Raman spectroscopy (LRS) has been previously shown to differentiate benign from malignant tissue in real time. However, the cost, assembly effort, and technical expertise needed for construction and implementation of the technique have prohibited widespread adoption. Recently, Raman spectrometers have been developed for non-medical uses and have become commercially available and affordable. Here we demonstrate that this current generation of Raman spectrometers can readily identify cancer in breast surgical specimens. We evaluated two commercially available, portable, near-infrared Raman systems operating at excitation wavelengths of either 785 nm or 1064 nm, collecting a total of 164 Raman spectra from cancerous, benign, and transitional regions of resected breast tissue from six patients undergoing mastectomy. The spectra were classified using standard multivariate statistical techniques. We identified a minimal set of spectral bands sufficient to reliably distinguish between healthy and malignant tissue using either the 1064 nm or 785 nm system. Our results indicate that current generation Raman spectrometers can be used as a rapid diagnostic technique distinguishing benign from malignant tissue during surgery.

Figure 1

Figure 1A shows the housing and Raman probe head common to both the i-Raman Plus (785 nm) and i-Raman Ex (1064 nm) systems. The housing measures 6.7″ × 13.4″ × 9.2” (17 cm × 34 cm × 23.4 cm), weighs ~10 lbs (4.6 kg) and is designed for operating temperatures between 10oC and 35oC. Figure 1B shows the collection of data from a surgical specimen in microscope mode with the Raman probe head integrated into the optical axis of a standard laboratory microscope. Figure 1C shows the Raman probe head in hand held mode encased in a sterile surgical sleeve.

Figure 2

Raw Raman spectra can distinguish healthy and neoplastic tissue. Figure 2A and B compare the fluorescence generated by the two systems. The average raw Raman spectra for healthy and neoplastic tissue samples acquired using 1064 nm (A) and 785 nm (B) excitation wavelengths are presented exactly as collected without smoothing, fluorescence correction or area normalization. Total laser exposure (defined as laser excitation power x collection time) was 9 × 10³ mW-seconds for both systems. Raman scattering data are reported in counts per second. The 1064 nm system exhibits less than half the fluorescence (A) generated by the 785 nm device (B). Fluorescence-corrected, normalized Raman spectra of healthy and neoplastic tissue following 785 nm and 1064 nm excitation appear in (C,D) and in (E), respectively. Full Raman shift spectra provided by the 785 nm device appear in (C). The strong Raman signal generated in the high wavenumber region by healthy tissue decreases significantly in the signals generated by malignant tissue. Comparison of tumor and healthy signals reveals a malignant spectral signature in normalized Raman spectra. Raman bands contributing to the signatures are marked graphically by gray bands and listed in Table 1 for both systems. (C,D) and (E) also exhibit a difference spectrum (gray line), highlighting the disparities between the average healthy and cancerous signatures. Positive deviations from neutral mark increased flux in tumor spectra, while negative deviations denote increased flux in healthy spectra. Due to the limited detector size of the 1064 nm system, the Raman spectrum high wavenumber region (2800–3200 cm⁻¹) can only be acquired using the 785 nm device.

Figure 3

PCA-LDA classification of Raman spectra generated by 1064 nm and 785 nm systems. Figure 3 depicts the PCA-LDA identification of two spectral classes for tissue regions that by visual morphological classification were either tumor-rich () or healthy (). We utilized the 3 PCA factors (Table 3) extracted from the 1064 nm and 785 nm data as inputs for LDA classification. Figure 3A is a plot of PC1 and PC2 factors extracted from 1064 nm spectral data from 57 targets in tissue regions that appeared either macroscopically healthy (N = 28) or tumor-rich (N = 29). LDA (Fig. 3B) classifies 27 of the 28 spectra from healthy regions as healthy, and 25 of 29 spectra from tumor-rich regions as pathological (sensitivity = 86%, specificity = 96%, and accuracy = 91%). Figure 3C is a plot of PC1 and PC2 factors extracted from 785 nm data from 50 targets in tissue regions that appeared either macroscopically healthy (N = 10) or tumor-rich (N = 40). LDA (Fig. 3D) classifies 10 of the 10 spectra from healthy regions as healthy, and 38 of 40 spectra from tumor-rich regions as pathological (sensitivity = 95%, specificity = 100%, and accuracy = 96%).

Figure 4

Average spectra for tissue classified as Healthy, Mixed, and Tumor by PCA-LDA. Spectra in (A) were generated using 1064 nm excitation. The figure depicts the average spectra for targets identified as Healthy and Tumor by PCA-LDA as well as the difference spectrum created by subtracting the Tumor (T) from the Healthy average spectra. The difference spectrum reveals increased flux for nucleic acid and protein bands at 941 and 1006 cm⁻¹ as well as loss of signal strength at 1271, 1063, 1080, 1303, 1442, and 1653 cm⁻¹ characteristic of shifts in protein and lipid species common in spectra from tumor-rich tissue. Spectra in (B) were acquired with the 785 nm system. Here the features in the difference spectrum distinguishing the Tumor from Healthy signatures are less pronounced than in the 1064 nm data, but once again there are increases in flux at 742, 941, and 1002 cm⁻¹ attributed to increased nucleotide and protein tissue concentrations as well as a marked loss of flux at 2850 attributed to a decrease in lipid content.

Figure 5

Representative tumor specimen and sites for collection of Raman spectra. The 785 system in microscope mode collected spectra along four transects designed to move sequentially across the visible boundaries between healthy and cancerous tissue. (A) Spectral collection sites along four transits pictured using a visible light image of an intact surgical specimen. Transit 1 sites are labelled 1–15; Transit 2 ran from 16 to 23; Transit 3 from 24 to 28; and Transit 4 from 29 to 37. Following data collection, the samples are fixed in formalin, embedded in paraffin, sectioned at 4 μm, and placed on glass slides for H&E staining. Whole slide images are obtained by scanning at 20X magnification. (B) The corresponding H&E stained section used for standard margin analysis is shown. Target sites are color-coded according to histological classification: green for healthy, red for regions dominated by tumor, and black for a mixture of healthy and cancerous tissue (“mixed”). (C) H&E photomicrographs for target sites s6 (green, healthy), s8 (black, mixed), and s11 (red, tumor) acquired during first transit depicted in (A) and (B). The Raman probe samples a circular area with a diameter of approximately 50–85 μm (in orange, a central spot representing the relative size of the laser beam). Refer to Fig. S1A for matching spectra.

Figure 6

Typical Changes in Raman Spectral Signatures during multiple data collection transits from healthy tissue to tumor tissue. A series of Raman spectra were obtained at ~1 mm intervals along a straight line moving from healthy to tumor tissue (or vice-versa). Such a series was termed a “transit”. By definition, each transit crosses the boundary between the two regions. Raman spectra for tissue sites along four transits are depicted. Each spectrum is the average of three scans, each with an integration time of 30 seconds. Total laser exposure time for each sample is 90 seconds. XY-coordinates for target location are recorded using the microscope micrometer. Spectra identifiers refer to target site designations depicted in Fig. 5. Spectra are numbered in temporal order of collection. For ease of viewing, spectra in Fig. 5 are offset and ordered (from top to bottom of the page) from data collected in putatively healthy tissue, across a boundary region, and then on into a tumor-rich region. For reference, the average spectra obtained from healthy (n = 88) and cancerous (n = 23) tissues, are depicted at the top and bottom, respectively, for each transit. Spectra s1-s5 were collected prior to first transit to evaluate signal/noise characteristics and are discussed in the supplemental material. Transit 1 starts with healthy spectra at sites s6 and s7. There is a clearly abnormal signature at s8, followed by a return to healthy spectra at sites s9 and s10. A clear shift to neoplastic spectra starts at s11 continuing through s15. Transit 2 starts with neoplastic signatures for sites s16-s19, then shifts to healthy spectra for sites s20-s23. Transit 3 traversed a region that appeared to be a mixture of tumor and healthy tissue in both the visible light and H&E images. All of the spectra (s24 through s28) appear to be a mixture of tumor and healthy signatures. Transit 4 starts in a tumor-rich region with spectra at s29-s32 closely resembling the average tumor spectra. The spectra then changes to a series of healthy tissue signatures at sites s33-s35, and finally shifts back to a tumor signature at sites s36 and s37.

Figure 7

Rapid characterization of tumor and healthy tissue using only the i-Ramani-Raman probe head without microscope. Spectral data were acquired with a single 10 second scan using the bare i-Ramani-Raman probe (785 nm excitation), uncoupled from the microscope, and secured in the probe holder. Spectral data shown are the averages of the 28 healthy and 29 tumor region spectra, where the fluorescence was corrected and the resulting spectra area normalized. Raman bands detecting significant activity from surgical marking ink are noted (693, 1260, 1348, 1398, 1541, and1597 cm⁻¹).

Figure 8

Tissue Classification Using Raman Spectra Collected Without Microscope. Figure 8 depicts the PCA-LDA identification of two spectral classes for tissue regions that by visual morphological classification were either tumor-rich () or healthy (). We utilized 3 PCA factors (Table 3) extracted from the 785 nm data as inputs for LDA classification. Fig. 8A is a plot of PC1 and PC2 factors extracted data from 57 targets in tissue regions that appeared either macroscopically healthy (N = 28) or tumor-rich (N = 29). LDA (B) classifies 24 of the 28 spectra from healthy regions as healthy, and 26 of 29 spectra from tumor-rich regions as pathological (sensitivity = 90%, specificity = 86%, and accuracy = 88%).