1.IntroductionBreast cancer is the most diagnosed cancer worldwide. It accounted for 1 in 8 cancer diagnoses and 685,000 deaths in 2020.1,2 It is the leading cause of cancer mortality in women and may affect men in rare cases.1,3 Early diagnosis and effective treatment are critical factors for survival.4 Ductal carcinoma is the most common type of breast cancer, and it originates in epithelial cells lining the interior of the ducts that carry milk from glands to the nipple. Lobular carcinoma, the second most common breast cancer type, originates from cells within the lobules, which are groups of glands connected to the duct. These carcinomas may be in situ or invasive.5 In situ cancers have not grown into surrounding tissues, whereas invasive cancer cells have initiated molecular processes allowing them to spread beyond the tissue of origin.5 There are two major types of in situ breast cancers: ductal carcinoma in situ and lobular carcinoma in situ.6 The most common breast imaging modality is x-ray mammography, generally followed by a biopsy and a histopathology analysis to assess cancer type.7 Adjuvant therapies are available to treat breast cancer (e.g., immunotherapy and chemotherapy) but standard of care involves surgical excision of the tumor followed by radiotherapy. Surgical options include breast-conserving surgery, mastectomy, and lymph nodes resection. Breast-conserving surgery is the recommended treatment for early stage patients. The aim of this procedure is to excise cancer with a margin of normal tissue surrounding the tumor. For early stage patients, long-term outcomes of breast-conserving surgery are equivalent to mastectomy if clear margins are obtained.8 The success of breast-conserving surgery therefore involves the complete removal of malignant tissue, including a circumferential margin of normal tissue surrounding the tumor. Because the presence of cancer cells on the margin is associated with a greater risk of cancer recurrence, specimen margins assessment by a pathologist can be required during a procedure to inform surgeons if more tissue needs to be removed. However, despite intraoperative pathological evaluation and radiological analyses, for up to 20% to 35% of cases positive margins are missed, resulting in the need for a second surgery.9,10 Re-excision procedures lead to additional costs, patient anxiety, and an increased risk of post-surgical complications, highlighting the need for new intraoperative surgical guidance techniques. Multiple emerging imaging techniques were developed for live in situ intra-operative breast tissue examination. Ultra-sound imaging was used to visualize structural tissue features and has proved an effective, rapid, and low-cost margins assessment technique.11 Radiofrequency spectroscopy, which is based on differences in scattering, reflectance, and absorbance, was designed for hand-held use and has shown promises for intraoperative margins detection.12 Impedance spectroscopy imaging, which relies on detecting dielectric tissue properties, was also developed as a portable handheld imaging device allowing rapid scanning.13 Although these techniques showed promise, breast cancer studies conducted using them showed levels of sensitivity and specificity that may limit their potential to reduce re-excision rates following breast-conserving surgery. Fiber-optics systems were also developed for intra-operative use relying on diffuse reflectance spectroscopy14 and optical coherence tomography.15 Other methods were used for ex vivo whole-specimen interrogation. Photoacoustic tomography used fat and hemoglobin as sources of contrast, leading to high-chemical selectivity but limited specificity to distinguish connective tissue and tumor.16 Spatial frequency domain imaging identified pathology subtypes based on the optical contrast provided by elastic scattering17 while other studies used either fluorescence lifetime imaging18 or tissue autofluorescence combined with diffuse reflectance spectroscopy.19 Microscopy techniques were developed for breast tissue studies with an emphasis on oncology. Those methods had a more limited field of view when compared to the in situ and ex vivo whole-specimen methods, but they were able to reveal more subtle structural and/or biochemical at high spatial resolution, with the enticing prospect to replace—or at least complement—standard histopathology analyses relying on staining techniques using dyes such as hematoxylin and eosin (H&E). A microscopic imaging approach based on ultra-violet surface excitation imaged tissue fluorophores without the requirement for tissue fixing, embedding, and sectioning.20 Light-sheet microscopy provided rapid, non-destructive, slide-free 2D and 3D imaging with the same level of detail as standard histopathology methods.21 Other microscopy techniques relied on non-linear optical contrast to provide high resolution molecular images.22–24 Raman spectroscopy showed potential for breast cancer detection. Macroscopic inelastic scattering point measurements25 and Raman micro-spectroscopy26,27 in biopsy specimens demonstrated the technique can be used to detect cancer. Raman spectroscopy studies by our group also showed potential for in situ intraoperative cancer detection of other pathologies, including brain,28–30 ovaries,31 and prostate.32–35 This work presents the results of a study that led to the acquisition of an ex vivo Raman spectroscopy dataset in human breast, including normal tissue and invasive ductal or lobular carcinoma. The study was designed to allow machine learning models to be developed based on the intrinsic macroscopic optical signature of multiple forms of tissue to distinguish invasive breast cancer from normal breast. The study also provided an assessment whether the classification models can shed light on specific biomolecular features of breast cancer using Raman spectroscopy, and whether the approach showed potential for surgical guidance in breast-conserving surgery to reduce re-excision procedure rates. A Raman spectroscopy single point probe system was used on fresh ex vivo specimens to train the cancer detection models but the intent is to use those same models for live in situ cancer detection during surgical procedures (Fig. 1). Fig. 1Schematic representation of the Raman spectroscopy hand-held probe workflow developed for surgical-guidance during breast-conserving surgery. Ex vivo spectroscopic measurements were made on breast surgery specimen to train machine learning models. The machine learning techniques selected the biomolecular spectral features with the most contrast between cancer and normal breast, and those features were used to train and test the cancer detection predictive models. The intent is to use those models for live in situ cancer detection during breast-conserving surgery. ![]() 2.Methods2.1.Raman Spectroscopy SystemMeasurements were made using an intra-operative hand-held single-point Raman spectroscopy probe system manufactured by the company Reveal Surgical (Montreal, Canada). The fiber-optics probe design of similar systems was presented elsewhere.30,36 Briefly, it integrates ten fiber-optics of diameter core. Of these fibers, nine were used for light detection and a central fiber was used for tissue excitation. A band-pass filter covered the tip of the excitation fiber, allowing illumination centered around 785 nm, and a high-pass filter was used for the collection fibers. The probe was connected to a 785-nm laser and a spectrometer with high sensitivity at wavelengths ranging from 800 to 900 nm, with an average resolution of across the spectral detection domain. A converging lens at the tip of the probe ensured contact measurements interrogated a spot size of diameter . The system was controlled by a proprietary software (Reveal Surgical, Montreal, Canada) that allowed acquisition parameters to be set by the user, including laser power, exposure time per spectrum, and number of repeated measurements (i.e., accumulations) at each point. 2.2.Patients SelectionAbout 20 patients were recruited for this study. Of those, 19 underwent breast surgery (lumpectomy or mastectomy) following a diagnosis of invasive breast cancer (lobular or ductal) and one patient underwent breast reduction surgery. For that patient, spectroscopic measurements were made from a breast in which no tumor was radiologically detectable. However, the patient was diagnosed with breast cancer associated with a tumor detected in the contralateral breast. All patients recruited in the study were undergoing breast surgery for the first time, did not have neoadjuvant therapy, had a cancer grade inferior to 4, and had a tumor larger than 1 cm. Patient specimens were utilized to build an ex vivo dataset of Raman spectroscopy measurements combined with histopathological and clinical data. Informed consent was obtained before the patient underwent surgery (McGill University Health Center Ethics Committees, approval number 2021-5997). Clinical data available included age, tumor type, tumor size, and Nottingham histologic score. Patient demographic details are provided in Table 1 (see also Table 3 for demographic details on each patients). The breast reduction surgery patient was recruited with the intention to acquire more measurements in healthy tissue to ensure a more balanced dataset. Table 1Clinical and pathological characteristics of all patients undergoing breast surgery (std: standard deviation; NA: not available; and IQR: interquartile range).
2.3.Specimen Handling and ex vivo Spectroscopic MeasurementsTwenty breast specimens (one per patient) were extracted by the surgeons (either S.M. or F.T.). The specimens were weighed, measured, and marked with ink, following institutional standards. They were then cut in serial slices of thickness. For each specimen, two smaller samples were cut from two different slices. Those samples were selected by the pathologist (A.O.) based on visual inspection with the objective of having one sample containing cancer cells and one sample mostly composed of healthy tissue. The samples were fixed on a cardboard support using two pins and a photograph was taken. The support was placed on a grid template and the Raman spectroscopy probe was fixed above the center of the grid [Fig. 2(a)]. Before each measurement, the cardboard support was moved to a different location on the template. The position with respect to the lower right corner of the cardboard support was recorded for each measurement. Fig. 2Spatial registration of Raman spectroscopy measurements with histopathology: (a) Representation of the technique where a fresh specimen was initially pinned to a cardboard support placed on a guiding template in the form of a grid. The circle superposed with the specimen represents the position of the probe above the grid. (b) End result of the methodology leading to the spatial registration of all spectroscopic measurements with an H&E-stained section. Different colors were used for some of the circles to improve visualization in cases when they overlapped. ![]() The laser power of the system was set at a value that resulted in 100 mW being delivered to the tissue surface. The number of repeat measurements per point (accumulations) was fixed at and the exposure time per spectrum ranged from 0.1 to 4 s, including dark noise (background) measurements made with the laser off. The exposure time was adjusted to ensure a raw signal intensity above 60% of the sensor dynamical range while avoiding saturation. All measurements were acquired within a one-hour timeframe after surgical excision and the size of the samples was limited to to preserve the integrity of the specimen and the tumor for margin evaluation, as part of the regular post-surgery clinical diagnostic workflow. The size limitation and the time constrain limited the number of independent Raman measurements to—on average—10 per sample (i.e., 20 per patient). After spectroscopic measurements, the samples were fixed in formalin, embedded in paraffin, and sectioned in 3 to histological sections. They were then stained with H&E as per institutional standards, resulting in one stained image for each sample. 2.4.Registration with Optical Measurements and Histology LABELSA four-step methodology was developed leading to spatial registration of all Raman spectroscopy measurements with corresponding locations on the H&E images using the Inkscape software (Inkscape’s Contributors). First, macroscopic photographs of the breast sample on the cardboard support [Fig. 2(a)] were taken. Second, the H&E images were superposed to the photograph and rotated to ensure the contours of the stained tissue sections matched as much as possible the contours seen on the photograph. Irregularities in the specimen shape were used as geometrical landmarks guiding this spatial registration process. Visible structures, including blood vessels and regions with high adipose content, were also used as tissue-based fiducial markers to ensure an accurate superposition of the photographs with the H&E images. Third, a numerical version of the registration grid was superposed to the H&E image, and it was used to pinpoint the location of every spectroscopic measurement. Fourth, each measurement location was annotated with a 3-mm diameter circle. The size of that circle was set to be two times larger when compared to the tip of the probe, accounting for potential spatial registration inaccuracies associated with sample deformations. As exemplified in Fig. 2(b), this led to images where colored circles indicated measurement locations superposed with the H&E images. Different colors were used for some of the circles to improve visualization in cases when they overlapped. The resulting images were analyzed by pathologists (D.T., F.A.) to determine which types of cells were contained within each circled region. Cells were classified into three categories: (1) cancer (tumor cells, tumor stroma, or necrosis); (2) normal: either normal breast (connective tissue, stroma, fibroblast, and collagen) or breast parenchyma (ducts and lobules); (3) fat (adipose cells). The number of cells from each category was counted and the percentage of cancer, normal, or fat cells within each circle were computed. A measurement was labeled as cancer if the percentage of cancer cells was equal or superior to 80%, normal if the percentage of normal cells was equal or superior to 80%, and fat if the percentage of adipose cells was equal or superior to 70%. All measurements that did not fit into one of the three classes were excluded. 2.5.Data Processing and LabelingThe data processing steps preceding the production of machine learning models led to the extraction of the inelastic scattering signature, for each measurement, from background contributions, including intrinsic tissue fluorescence.37 Spectral pre-processing included the following steps: (1) averaging of the repeat spectra to increase overall signal-to-noise ratio, (2) averaging and subtraction of background spectra acquired with the laser turned off between measurements, (3) normalization with a NIST Raman standard (SRM 2214) to correct for the instrument response, (4) -axis (wavenumber shift) calibration based on a Raman spectrum acquired on polycarbonate, (5) removal of background signals using the custom background removal algorithm BubbleFill,38 and (6) standard normal variate (SNV) normalization. A quantitative quality factor metric (between 0 and 1) was then computed from each resulting spectrum. It provided a statistical assessment of the likelihood the SNV-normalized signal is associated with tissue Raman peaks or stochastic noise.39 Spectra with a quality factor metric inferior to 0.6 were excluded as they did not always clearly show all the ubiquitous Raman peaks encountered in biological tissue (e.g., amide bands for protein-rich tissue). All spectra were then visually inspected individually to qualitatively identify any glaring labeling errors here, mostly based on the fact that adipose tissue Raman spectra have spectral features that are dramatically different when compared to protein-rich tissue. For example, pure fat does not contain phenylalanine and as a result does not present a visually detectable peak at . Spectra labeled as cancer or normal that were associated with clearly identifiable fat spectra were removed from the dataset, as well as spectra identified as fat that did not show the expected biomolecular features associated with proteins. All remaining spectra were then plotted along with the standard deviation for each of the following groups: normal, cancer, and fat. The average spectrum associated with the normal breast measurements of the breast reduction surgery patient was also plotted against all other normal spectra and the cancer spectra (Fig. 7). 2.7.Machine Learning WorkflowA machine learning workflow was applied to develop three different models: cancer versus normal + fat (model A), cancer versus normal (model B), and cancer versus fat (model C). Due to the small size of the dataset and to minimize the risk of over-fitting the data, the first step of the machine learning workflow consisted of dimensionally reducing the number of features per spectra to for each model. This dimensional reduction procedure was done using a linear support vector machine (SVM) with a regularization by Lasso regression, in which individual statistical weights were assigned to each feature. This statistical weight acted as a surrogate for feature relevance, in effect quantifying the ability of each feature to capture inter-class variations and allowed less important bands to be discarded.40,41 Features were then ranked according to their weight and boxplots were produced to show the statistical distribution of spectral intensity for all features that were retained to produce the machine learning models (Table 2 and Fig. 5). Table 2Raman peaks and main corresponding vibrational bonds associated with the Raman-predicted molecular tissue content associated with the spectra in Fig. 3. The bands associated with spectral features used as inputs to the classification models (Model A, Model B, and Model C) are identified in the last column. A tentative molecular assignment (specific molecules and families of biomolecules) based on literature findings is shown.42
C─H: carbon-hydrogen single bonds, C═C: carbon-carbon double bonds (unsaturated), C─C: carbon-carbon bonds, CH2: ethyl group, CH3: methyl group. The classification models training process involved SVM with a linear kernel and a regularization parameter . This hyperparameter, which varied between and 1, controlled the penalty for errors in the training process and reduced the risk of overfitting the data. The SVM algorithm also considered the class imbalance for all datasets by setting a higher cost to misclassified spectra from the least populated class.41 A grid search method was used to optimize both hyperparameters and , and a five-fold cross-validation procedure was performed to assess classification performances. The five-fold cross validation took the training dataset and randomly split it in five equal-size groups. Then, five models were created with the same hyperparameters. Each model was trained on four groups and tested on the remaining one. The average of the performance of the five classifiers performance was used to assess predictive performance.43 During the cross-validation phase, data from the same patient were never used in both the training and testing sets. The technique returned, for each measurement, a posterior probability that a measurement was classified within one of two classes. A receiver-operating-characteristic (ROC) curve was computed by comparing this posterior probability with a parameter . All observations associated by the classifier were assigned the label 1 when or assigned the label 0 if otherwise. Different values of the parameter corresponded to different points of the ROC curve. Sensitivity and specificity were computed by comparing the label assigned by the model to the label given by the histopathology analysis for every measurement. Results were reported for the optimized hyperparameters only, yielding the highest area-under-curve (AUC) value. The final optimized model, with reported accuracy, sensitivity, and specificity, corresponded to the ROC curve point with the smallest distance to the upper left corner of the curve (-axis: 1 − specificity, -axis: sensitivity). 3.Results3.1.Spectroscopic MeasurementsApplication of the Raman spectroscopy measurement protocol resulted in 388 spectroscopic measurements with co-located histopathology analyses. Of those, 58 spectra were excluded because of visibly low spectral quality, non-tissue-related artifacts (e.g., from residual ambient light, Raman peaks from the cardboard substrate) or a quality factor lower than 0.6. Moreover, 58 measurements were excluded because they did not qualify for association with either the normal, cancer, or fat categories. Further, 34 spectra were excluded due to suspected labeling errors based on visual inspection comparing measured spectra with the known Raman signature of adipose tissue. Of those, 28 that were labeled as normal or cancer were excluded because the spectra were identical to fat spectra. The remaining six spectra, which were labeled as fat, were excluded because they showed clearly visible spectral features associated with protein-rich tissue. The final dataset was composed of 238 measurements: 58 were labeled as normal, 87 as cancer, and 93 as fat. The raw and processed average measurements for each tissue category, along with the signal variance associated with each spectral bin, are shown in Figs. 3(a) and 3(b), respectively. Selected bands of interest are identified in Fig. 3(b) and their biochemical interpretation is provided in Table 2. Fig. 3Spectra associated with normal, cancer, and fat measurements in breast tissue: (a) average spectroscopic measurements (raw data) for each tissue category along with the standard deviation for each spectral bin; (b) average Raman spectra and standard deviation with selected bands identified with their wavenumber value. All identified bands were used for the biochemical interpretation of the data (Table 2), and a subset of those bands was used for machine learning model development. ![]() 3.2.Machine Learning Models and Biomolecular PredictionsA classification model (Model A) was trained to distinguish between the cancer and non-cancer (fat + normal) spectra, from 238 measurements (87 cancer spectra, 58 normal spectra, and 93 fat spectra). This yielded an accuracy of 94%, a sensitivity of 93% and a specificity of 95%. The ROC curve for this model is shown in Fig. 4(a) and it is associated with an AUC of 0.98. These results were achieved using a model trained using only two spectral features, namely the peaks mostly associated with C–C stretching of proteins around and the symmetric aromatic ring breathing at that is associated with the presence of phenylalanine. A classification model (Model B) was also trained to distinguish cancer from normal spectra from 145 measurements (87 cancer spectra and 58 normal spectra). This model had an accuracy of 91%, a sensitivity of 92%, a specificity of 90%, and an AUC of 0.96 based on four spectral features [Fig. 4(b)]. Those features corresponded to the same features as Model A, plus a peak associated with C–C skeletal stretching from proteins, lipids, and carbohydrates () and a band around associated with C-C/C-N skeletal stretching from proteins. Model C was developed to detect cancer from fat based on 180 Raman measurements (87 cancer spectra and 93 fat spectra). It performed with an accuracy of 98%, a sensitivity of 97%, and a specificity of 98%, with an AUC of almost 1 [Fig. 4(c)]. This model required two spectral features, namely, the peak and the twisting deformation band at 1301 to . Fig. 4Average Raman spectra and spectral features (dotted lines) used by the classification models. Classification results are expressed in term of accuracy, sensitivity, and specificity. Confusion matrices and ROC curves are also shown for the three models: (a) Model A: cancer versus non-cancer (normal + fat); (b) Model B: cancer versus normal, and (c) Model C: cancer versus fat. ![]() The spectral features that were used by the machine learning models are given in Table 2 along with their associated vibrational bonds and a tentative assignment with biomolecules based on the literature. Figure 5 shows the statistical distribution of these bands in the form of boxplots. The median (second quartile, Q2), first quartile (Q1), and third quartile (Q3) were clearly distinct for the two features used in model A (940 and ). Q1, Q2, and Q3 were also well separated for cancer and normal tissue features associated with model B (940, 1004, 1129, and ) as well as for cancer and fat for the features used in model C (1004 and ). Fig. 5Statistical distribution of the intensity of the five features used by classification models for fat, normal and cancer spectra. Model A (cancer versus normal + fat) used features associated with bands at 940 and ; Model B (cancer versus normal) used bands at 940, 1004, 1129, and ; and Model C (cancer versus fat) used bands at 1004 and . Outliers are represented by black diamonds. ![]() A fourth model was trained to distinguish fat from all other tissue types (normal + cancer) from 238 spectra (Fig. 6), resulting in an almost perfect AUC of 1 with specificity and sensitivity . This model was used to test whether the exclusion criteria leading to the rejection of suspected mis-labeled samples (28 labeled as normal or cancer, 6 spectra labeled as fat) was justified. The model was directly applied to those 34 spectra and predicted, in all cases, that the histology labels were inconsistent with the measured Raman signature. Specifically, the model predicted that all 28 normal or cancer spectra were actually fat, and that all 6 fat spectra were actually normal or cancer. Fig. 6A Fat detection classifier was trained from 238 spectra. It was applied on the 34 spectra that were excluded from the study because of suspected mis-labeling likely associated with spatial registration errors of Raman spectroscopy measurements with the H&E images: the 28 spectra that were labeled as normal or cancer were all classified as fat (false negatives) and the six spectra that were labeled as fat were classified as normal or cancer (false positives). ![]() In this study, more than 30% of all normal measurements were associated with the breast conserving surgery patient, which could have led to classification biases. However, Fig. 7 qualitatively shows that the average normal spectrum from the breast conserving surgery patient ( spectra) was similar to the average spectrum associated to all other normal spectra in the dataset (). More importantly, the average spectra associated with the normal category matched perfectly for the Raman bands used to train Models A, B, and C (Table 2), providing strong evidence no bias resulted from including a larger proportion of normal measurements from the same patient. 4.DiscussionThis work demonstrated that it is possible to develop low complexity machine learning models with the ability to distinguish invasive breast cancer from healthy tissue. Here, the term low complexity means that all machine learning models were trained using only a few spectral features, each associated with known biomolecular vibrational modes. The resulting models are thus easily interpretable, a trait that could lower barriers to clinical adoption. This is important since the use of a small number of features reduces the burden imposed by the requirement for large datasets, a common issue encountered in biomedical applications of machine learning. Indeed, the confidence in estimated model performances achieved through cross-validation approaches is always higher when the number of features used during the model training phase is significantly smaller when compared to the number of training instances. In this work, 238 measurements from 20 patients were available for predictive modeling, and spectral bands were used. This meant that the number of data types used for learning was lower than 5%, therefore much smaller than rule-of-thumb values recommended in machine learning, i.e., usually around 20% to 25%. Another important aspect of this work was that significant efforts were made to ensure the trends associated with all retained Raman spectral features reflected—with high confidence—the actual nature of the tissue. This was achieved by ensuring model training and validation were done only using Raman measurements associated with high spectral quality and for which the spatial registration with histology analysis could be trusted. In fact, data selection was done to mitigate the limited reliability associated with the gold standard of histology used to assign labels to each measurement. For those data points excluded from the machine learning model development process, the registration of the Raman measurements with the H&E-stained images resulted in labeling errors when assigning spectra to either the cancer, normal, or fat category. One cause of error was the change in shape between the specimen photograph and the H&E-stained image of the corresponding tissue section. The samples were compressed and rotated before being processed and stained, which sometimes made superposing the photograph with the H&E images difficult. Also, the specimen thickness was 1 cm or more and was sectioned in a 3 to section before staining. Therefore, the surface on which the Raman measurements were acquired could be different than that located on the stained tissue sections. This source of inaccuracy was mitigated by having the location of each measurement marked on the H&E-stained image by a 3-mm diameter area that was significantly larger than the tissue area sampled with the Raman spectroscopy probe. Data points where the number of cells from the same category was too low (cancer: , normal: , and fat: ) within this 3-mm diameter area were also excluded. A more reliable spatial registration method between the gold standard of histology and the measurements would highly reduce the number of non-usable spectra that needed to be excluded in this study. Using well-justified data point rejection criteria led to a smaller dataset, albeit associated with spectra having low inter-patient and intra-specimen variance. This allowed a clear biomolecular interpretation of the key Raman-predicted biomolecular differentiators between healthy and cancer tissue. The only feature present in all three models was the peak corresponding to the aromatic amino acid phenylalanine. Other groups such as Zuniga et al. also obtained a sharp peak at the same Raman shift in breast cancer tissue.44 Compared to normal and adipose tissues, cancerous tissues had an increase of intensity at the phenylalanine peak. This was also confirmed by the increase of the peak at , which could also be associated with phenylalanine [Fig. 4(b)]. Gebrekidan et al. and Contorno et al. also found similar results of higher content in phenylalanine in breast tumor tissues.45,46 They hypothesized that this increase could be caused by the fact phenylalanine is the precursor of substances with mutagenic, genotoxic, and carcinogenic properties and that overexpression may be suggestive of the cancer pathogenesis. The DNA/RNA peaks at 785, 1176, , and the proteins peaks at 760 and were higher in breast tumor tissue. Many studies have confirmed genetic alterations in breast cancer disease.47,48 Specifically, the peaks at 785 and correspond to vibrational modes of cytosine and several studies have found the implication of spontaneous and enzyme-catalyzed deamination of DNA cytosine bases.49 Similar findings of enhanced peaks of Amide III (peaks at 1246 and ) have been reported in a study by Liu et al.50 Interestingly, Bitar et al.51 found that there were differences in the secondary structure of proteins of breast tissues that could differentiate subtypes of malignant lesions. In addition, they found that collagen content could be used to quantify the degree of pathogenicity of breast cancer. The two peaks at 1129 and corresponding to lipids were also enhanced in the breast cancer tissues compared to normal breast tissues. However, during the past years, our laboratory work in other pathologies has shown that most tumors have a lower intensity for peaks associated with lipids compared to normal tissues.32,52 Lipid metabolism in tumor cells has been widely studied, their biosynthesis is tightly related to signaling pathway. Li et al. found a decrease in lipids and suggested that tumors share a common phenotype of uncontrolled cell proliferation and, therefore, they consume large amounts of lipids.53 However, more recent studies have found a specific type of increase in lipids in breast tumor tissues: the unsaturated fatty acid. In 2016, Sixian You et al. observed an increase of polyunsaturated fatty acids in the human breast cancer tissues.54 Brozek-Pluska et al. (2019) also confirms the poly unsaturated fatty acid important increase in these malignant tissues.55 They have cautioned that disturbed ratio content of unsaturated and saturated lipids seems to be the marker of pathogenesis and not only the increase of polysaturated fatty acid. Further investigations would be needed to determine the different Raman spectra of polyunsaturated and monosaturated fatty acids composition in breast cancer. Another tentative hypothesis is that while the band at is solely associated with lipids, the band of is however assigned to stretching (proteins), stretching (carbohydrates), and lipid transconformation. The contribution of proteins in cancer tissues adding to the presence of carbohydrates of the band might have contributed to the upshifting of . All in all, lipids band assignments should be taken cautiously in breast cancer tissues. While there are contradictory findings about the increase or decrease of lipids in literature and among Raman spectroscopy work in breast cancer, the future work should be focusing on clarifying the roles of the lipids contribution in its malignancy. One way to do it would be to study the lipid phenotype of breast cancer cells and its metastatic spread. For example, Nieva et al.56 reported in 2014 the ability of Raman micro-spectroscopy to achieve rapid lipid profiling of breast cancer cells and therefore, distinguish metastatic ability from malignancy. The intended use of the Raman spectroscopy system presented in this study is live surgical guidance during breast-conserving surgery. However, the total measurement times that were reported ranged from 1 to 40 s, for repeat measurements. Such large tissue interrogation times could impact practical use during surgical procedures. However, the data pre-processing and machine learning workflow used to produce the predictive models were re-applied on the subset of measurements associated with and accumulations, to assess potential degradation in performance associated with limited interrogation times. Interestingly, the accuracy and AUC of the ROC curve for those new models were essentially unchanged for , and they were associated with limited changes for . For example, for the accuracy and AUC of the cancer versus normal model was 88% and 0.89, compared to 91% and 0.97 for . This implies that high levels of predictive accuracy could still be reached when lowering the range of measurement times to 0.1 to 4 s rather than 1 to 40 s. Other practical concerns relate with the maximal permissible exposure (MPE) to tissue as set by ANSI standards for research laboratory laser safety.57 As an example, for a total light exposure duration of 1 s, MPE for skin is ; for total duration , as was the case for the highest total integration time of 40 s, the MPE for skin is . In comparison, the measurements in this work led to energy densities of when computed for hazard evaluation (i.e., 3.5-mm diameter limiting aperture) based on the rules set forth in ANSI Z136.1-2014 American National Standard for Safe Use of Lasers. It should be noted that those skin MPE values do not correspond to absolute thresholds beyond which damage to tissue will necessarily result in breast tissue. For example, none of the specimens in which Raman spectroscopy measurements were made presented morphological alterations (as assessed based on histology analyses) suggestive of heat-generated tissue damage. Future clinical applications using Raman spectroscopy in breast-conserving surgery will require detailed analyses to be done to determine acceptable photo-diagnostic levels, which will likely be significantly larger than the current MPE values set for skin. Although the results presented here are very promising, work needs to be done to bring Raman spectroscopy in operating rooms for breast-conserving surgery guidance. For example, single-point measurements may not provide information sufficiently rapidly for surgeons to influence intraoperative decisions. This is because inspection of surgical cavities with large surface areas are expected in most cases, making successive single points of millimeter-sized areas unlikely to make the approach clinically deployable. To solve this problem, our group developed a handheld macroscopic Raman spectroscopy imaging instrument, specifically adapted for surgical guidance applications in breast oncology58,59 This system demonstrated hyperspectral detection of the fingerprint Raman signal over an area of , and classification model transferability across different biomedical Raman spectroscopy instruments was demonstrated, included with the Raman imaging system and the single-point probes.32 In the future, the classification models built in this study will therefore be used with the Raman imaging system for further studies in breast cancer.59 5.Appendix: Supplemental MaterialTable 3 presented clinical information such as age, type of surgery, and type of cancer for every patient included in the study. It also shows the number of measurements taken on each class of tissues for each patient. Table 3Demographic and clinical information of the 20 patients included in the study.
DisclosureFrederic Leblond is co-founder of ODS Medical (now Reveal Surgical) formed in 2015 to commercialize a Raman spectroscopy system for neurosurgical and prostate surgery applications. He has ownership and patents in the company. AcknowledgmentsThis work is supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) Alliance grant program, Reveal Surgical, the NSERC Discovery Grant program and the TransMedTech Institute. SD was supported by NSERC. DT receives salary support from the FRQS (Clinical Research Scholar, Junior 2). The CRCHUM receives support from the FRQS. Access to Dr. Feryel Azzi’s expertise is possible thanks to the TransMedTech Institute and its primary funding partner, the Canada First Research Excellence Fund. Code, Data, and Materials AvailabilityThe data and materials information that support the findings of this study are available from the corresponding author upon reasonable request. ReferencesH. Sung et al.,
“Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries,”
CA. Cancer J. Clin., 71
(3), 209
–249 https://doi.org/10.3322/caac.21660 CAMCAM 0007-9235
(2021).
Google Scholar
M. Arnold et al.,
“Current and future burden of breast cancer: global statistics for 2020 and 2040,”
Breast, 66
(June), 15
–23 https://doi.org/10.1016/j.breast.2022.08.010
(2022).
Google Scholar
A. U. Buzdar,
“Breast cancer in men,”
Oncology, 17
(10), 1361
–1364 https://doi.org/10.1177/1557988311416495
(2003).
Google Scholar
P. M. Webb et al.,
“Changes in survival after breast cancer: improvements in diagnosis or treatment?,”
Breast, 13
(1), 7
–14 https://doi.org/10.1016/S0960-9776(03)00129-2
(2004).
Google Scholar
B. J. Yoder, E. J. Wilkinson and N. A. Massoll,
“Molecular and morphologic distinctions between infiltrating ductal and lobular carcinoma of the breast,”
Breast Journal, 13 172
–179 https://doi.org/10.1111/j.1524-4741.2007.00393.x
(2007).
Google Scholar
E. M. Ward et al.,
“Cancer statistics: breast cancer in situ,”
CA. Cancer J. Clin., 65 481
–495 https://doi.org/10.3322/caac.21321 CAMCAM 0007-9235
(2015).
Google Scholar
L. Bartella et al.,
“Imaging breast cancer,”
Radiol. Clin. North Am., 45
(1), 45
–67 https://doi.org/10.1016/j.rcl.2006.10.007
(2007).
Google Scholar
A. A. Onitilo et al.,
“Survival comparisons for breast conserving surgery and mastectomy revisited: community experience and the role of radiation therapy,”
Clin. Med. Res., 13
(2), 65
–73 https://doi.org/10.3121/cmr.2014.1245
(2015).
Google Scholar
F. J. Fleming et al.,
“Intraoperative margin assessment and re-excision rate in breast conserving surgery,”
Eur. J. Surg. Oncol., 30 233
–237 https://doi.org/10.1016/j.ejso.2003.11.008
(2004).
Google Scholar
L. Jacobs,
“Positive margins: the challenge continues for breast surgeons,”
Ann. Surg. Oncol., 15
(5), 1271
–1272 https://doi.org/10.1245/s10434-007-9766-0
(2008).
Google Scholar
M. Moschetta et al.,
“Role of specimen us for predicting resection margin status in breast conserving therapy,”
G. di Chir., 36 201
–204 https://doi.org/10.11138/gchir/2015.36.5.201
(2015).
Google Scholar
M. Thill,
“MarginProbe®: intraoperative margin assessment during breast conserving surgery by using radiofrequency spectroscopy,”
Expert Rev. Med. Devices, 10 301
–315 https://doi.org/10.1586/erd.13.5 1743-4440
(2013).
Google Scholar
J. M. Dixon et al.,
“Intra-operative assessment of excised breast tumour margins using ClearEdge imaging device,”
Eur. J. Surg. Oncol., 42 1834
–1840 https://doi.org/10.1016/j.ejso.2016.07.141
(2019).
Google Scholar
I. J. Bigio et al.,
“Diagnosis of breast cancer using elastic-scattering spectroscopy: preliminary clinical results,”
J. Biomed. Opt., 5
(2), 221
–228 https://doi.org/10.1117/1.429990 JBOPFO 1083-3668
(2000).
Google Scholar
S. J. Erickson-Bhatt et al.,
“Real-time imaging of the resection bed using a handheld probe to reduce incidence of microscopic positive margins in cancer surgery,”
Cancer Res., 75 3706
–3712 https://doi.org/10.1158/0008-5472.CAN-15-0464 CNREA8 0008-5472
(2015).
Google Scholar
R. Li et al.,
“Assessing breast tumor margin by multispectral photoacoustic tomography,”
Biomed. Opt. Express, 6 1273
–1281 https://doi.org/10.1364/BOE.6.001273 BOEICL 2156-7085
(2015).
Google Scholar
A. M. Laughney et al.,
“Spectral discrimination of breast pathologies in situ using spatial frequency domain imaging,”
Breast Cancer Res., 15 R61 https://doi.org/10.1186/bcr3455 BCTRD6
(2013).
Google Scholar
J. E. Phipps et al.,
“Automated detection of breast cancer in resected specimens with fluorescence lifetime imaging,”
Phys. Med. Biol., 63 015003 https://doi.org/10.1088/1361-6560/aa983a PHMBA7 0031-9155
(2018).
Google Scholar
M. D. Keller et al.,
“Autofluorescence and diffuse reflectance spectroscopy and spectral imaging for breast surgical margin analysis,”
Lasers Surg. Med., 42 15
–23 https://doi.org/10.1002/lsm.20865 LSMEDI 0196-8092
(2010).
Google Scholar
F. Fereidouni et al.,
“Microscopy with ultraviolet surface excitation for rapid slide-free histology,”
Nat. Biomed. Eng., 1 957
–966 https://doi.org/10.1038/s41551-017-0165-y
(2017).
Google Scholar
A. K. Glaser et al.,
“Light-sheet microscopy for slide-free non-destructive pathology of large clinical specimens,”
Nat. Biomed. Eng., 1 0084 https://doi.org/10.1038/s41551-017-0084
(2017).
Google Scholar
Y. K. Tao et al.,
“Assessment of breast pathologies using nonlinear microscopy,”
Proc. Natl. Acad. Sci. U. S. A., 111 15304
–15309 https://doi.org/10.1073/pnas.1416955111
(2014).
Google Scholar
J. Mertz,
“Nonlinear microscopy: new techniques and applications,”
Current Opinion in Neurobiology, 14 610
–616 https://doi.org/10.1016/j.conb.2004.08.013 COPUEN 0959-4388
(2004).
Google Scholar
Y. Yang et al.,
“Label-free imaging of human breast tissues using coherent anti-Stokes Raman scattering microscopy,”
Proc. SPIE, 7903 79032G https://doi.org/10.1117/12.871063 PSISDG 0277-786X
(2011).
Google Scholar
A. S. Haka et al.,
“Diagnosing breast cancer by using Raman spectroscopy,”
Proc. Natl. Acad. Sci. U. S. A., 102
(35), 12371
–12376 https://doi.org/10.1073/pnas.0501390102
(2005).
Google Scholar
K. Kong et al.,
“Towards intra-operative diagnosis of tumours during breast conserving surgery by selective-sampling Raman micro-spectroscopy,”
Phys. Med. Biol., 59 6141
–6152 https://doi.org/10.1088/0031-9155/59/20/6141 PHMBA7 0031-9155
(2014).
Google Scholar
H. Abramczyk et al.,
“Raman ‘optical biopsy’ of human breast cancer,”
Prog. Biophys. Mol. Biol., 108
(1–2), 74
–81 https://doi.org/10.1016/j.pbiomolbio.2011.10.004 PBIMAC 0079-6107
(2012).
Google Scholar
É. Lemoine et al.,
“Feature engineering applied to intraoperative: in vivo Raman spectroscopy sheds light on molecular processes in brain cancer: a retrospective study of 65 patients,”
Analyst, 144 6517
–6532 https://doi.org/10.1039/C9AN01144G ANLYAG 0365-4885
(2019).
Google Scholar
M. Jermyn et al.,
“Highly accurate detection of cancer in situ with intraoperative, label-free, multimodal optical spectroscopy,”
Cancer Res., 77 3942
–3950 https://doi.org/10.1158/0008-5472.CAN-17-0668 CNREA8 0008-5472
(2017).
Google Scholar
M. Jermyn et al.,
“Intraoperative brain cancer detection with Raman spectroscopy in humans,”
Sci. Transl. Med., 7 274ra19 https://doi.org/10.1126/scitranslmed.aaa2384 STMCBQ 1946-6234
(2015).
Google Scholar
S. David et al.,
“Multispectral label-free Raman spectroscopy can detect ovarian and endometrial cancer with high accuracy,”
J. Biophotonics, 15
(2), 1
–10 https://doi.org/10.1002/jbio.202100198
(2022).
Google Scholar
F. Picot et al.,
“Image-guided Raman spectroscopy navigation system to improve transperineal prostate cancer detection. Part 1: Raman spectroscopy fiber-optics system and in situ tissue characterization,”
J. Biomed. Opt., 27
(9), 095003 https://doi.org/10.1117/1.JBO.27.9.095003 JBOPFO 1083-3668
(2022).
Google Scholar
D. Grajales et al.,
“Image-guided Raman spectroscopy navigation system to improve transperineal prostate cancer detection. Part 2: In-vivo tumor-targeting using a classification model combining spectral and MRI-radiomics features,”
J. Biomed. Opt., 27
(9), 095004 https://doi.org/10.1117/1.JBO.27.9.095004 JBOPFO 1083-3668
(2022).
Google Scholar
K. Aubertin et al.,
“Combining high wavenumber and fingerprint Raman spectroscopy for the detection of prostate cancer during radical prostatectomy,”
Biomed. Opt. Express, 9 4294
–4305 https://doi.org/10.1364/BOE.9.004294 BOEICL 2156-7085
(2018).
Google Scholar
K. Aubertin et al.,
“Mesoscopic characterization of prostate cancer using Raman spectroscopy: potential for diagnostics and therapeutics,”
BJU Int., 122 326
–336 https://doi.org/10.1111/bju.14199 BJINFO 1464-410X
(2018).
Google Scholar
J. Desroches et al.,
“Development and first in-human use of a Raman spectroscopy guidance system integrated with a brain biopsy needle,”
Journal of Biophotonics, 12 e201800396 https://doi.org/10.1002/jbio.201800396
(2019).
Google Scholar
D. Wei, S. Chen and Q. Liu,
“Review of fluorescence suppression techniques in Raman spectroscopy,”
Applied Spectroscopy Reviews, 50 387
–406 https://doi.org/10.1080/05704928.2014.999936 APSRBB 0570-4928
(2015).
Google Scholar
G. Sheehy et al.,
“Open-sourced Raman spectroscopy data processing package implementing a novel baseline removal algorithm validated from multiple datasets acquired in human tissue and biofluids,”
J. Biomed. Opt., 28
(2), 025002 https://doi.org/10.1117/1.JBO.28.2.025002 JBOPFO 1083-3668
(2023).
Google Scholar
F. Dallaire et al.,
“Quantitative spectral quality assessment technique validated using intraoperative in vivo Raman spectroscopy measurements,”
J. Biomed. Opt., 25
(4), 040501 https://doi.org/10.1117/1.JBO.25.4.040501 JBOPFO 1083-3668
(2020).
Google Scholar
D. M. Hawkins,
“The problem of overfitting,”
J. Chem. Inf. Comput. Sci., 44
(1), 1
–12 https://doi.org/10.1021/ci0342472 JCISD8 0095-2338
(2004).
Google Scholar
J. Neumann, C. Schnörr and G. Steidl,
“Combined SVM-based feature selection and classification,”
Mach. Learn., 61
(1–3), 129
–150 https://doi.org/10.1007/s10994-005-1505-9 MALEEZ 0885-6125
(2005).
Google Scholar
A. C. S. Talari et al.,
“Raman spectroscopy of biological tissues,”
Appl. Spectrosc. Rev., 50
(1), 46
–111 https://doi.org/10.1080/05704928.2014.923902 APSRBB 0570-4928
(2015).
Google Scholar
S. S. Skiena, The Data Science Design Manual, Springer(
(2017). Google Scholar
W. C. Zúñiga et al.,
“Raman spectroscopy for rapid evaluation of surgical margins during breast cancer lumpectomy,”
Sci. Rep., 9 14639 https://doi.org/10.1038/s41598-019-51112-0 SRCEC3 2045-2322
(2019).
Google Scholar
M. T. Gebrekidan et al.,
“A shifted-excitation Raman difference spectroscopy (SERDS) evaluation strategy for the efficient isolation of Raman spectra from extreme fluorescence interference,”
J. Raman Spectrosc., 47
(2), 198
–209 https://doi.org/10.1002/jrs.4775 JRSPAF 0377-0486
(2016).
Google Scholar
S. Contorno, R. E. Darienzo and R. Tannenbaum,
“Evaluation of aromatic amino acids as potential biomarkers in breast cancer by Raman spectroscopy analysis,”
Sci. Rep., 11 1698 https://doi.org/10.1038/s41598-021-81296-3 SRCEC3 2045-2322
(2021).
Google Scholar
C. Swanton et al.,
“APOBEC enzymes: mutagenic fuel for cancer evolution and heterogeneity,”
Cancer Discov., 5
(7), 704
–712 https://doi.org/10.1158/2159-8290.CD-15-0344
(2015).
Google Scholar
S. Nik-Zainal et al.,
“Mutational processes molding the genomes of 21 breast cancers,”
Cell, 149
(5), 979
–993 https://doi.org/10.1016/j.cell.2012.04.024 CELLB5 0092-8674
(2012).
Google Scholar
B. J. M. Taylor et al.,
“DNA deaminases induce break-associated mutation showers with implication of APOBEC3B and 3A in breast cancer kataegis,”
Elife, 2013
(2), 1
–14 https://doi.org/10.7554/eLife.00534
(2013).
Google Scholar
C. H. Liu et al.,
“Resonance Raman and Raman spectroscopy for breast cancer detection,”
Technol. Cancer Res. Treat., 12
(4), 371
–382 https://doi.org/10.7785/tcrt.2012.500325
(2013).
Google Scholar
R. A. Bitar et al.,
“Biochemical analysis of human breast tissues using Fourier-transform Raman spectroscopy,”
J. Biomed. Opt., 11
(5), 054001 https://doi.org/10.1117/1.2363362 JBOPFO 1083-3668
(2006).
Google Scholar
A. A. Grosset et al.,
“Identification of intraductal carcinoma of the prostate on tissue specimens using Raman micro-spectroscopy: a diagnostic accuracy case-control study with multicohort validation,”
PLoS Med., 17
(8), e1003281 https://doi.org/10.1371/journal.pmed.1003281
(2020).
Google Scholar
H. Li et al.,
“Raman microspectroscopic investigation and classification of breast cancer pathological characteristics,”
Molecules, 26
(4), 921 https://doi.org/10.3390/molecules26040921
(2021).
Google Scholar
S. You et al.,
“Raman spectroscopic analysis reveals abnormal fatty acid composition in tumor micro- and macroenvironments in human breast and rat mammary cancer,”
Sci. Rep., 6
(Dec. 2015), 1
–10 https://doi.org/10.1038/srep32922 SRCEC3 2045-2322
(2016).
Google Scholar
B. Brozek-Pluska et al.,
“Label-free diagnostics and cancer surgery Raman spectra guidance for the human colon at different excitation wavelengths,”
RSC Adv., 9
(69), 40445
–40454 https://doi.org/10.1039/C9RA06831G
(2019).
Google Scholar
C. Nieva et al.,
“The lipid phenotype of breast cancer cells characterized by Raman microspectroscopy: towards a stratification of malignancy,”
PLoS One, 7
(10), e46456 https://doi.org/10.1371/journal.pone.0046456 POLNCL 1932-6203
(2012).
Google Scholar
B. Sams,
“American National Standard,”
SMPTE J., 95
(5), 601
–602 https://doi.org/10.5594/j17740 SMPJDF 0036-1682
(2012).
Google Scholar
F. Daoust et al.,
“Handheld macroscopic Raman spectroscopy imaging instrument for machine-learning-based molecular tissue margins characterization,”
J. Biomed. Opt., 26
(2), 022911 https://doi.org/10.1117/1.JBO.26.2.022911 JBOPFO 1083-3668
(2021).
Google Scholar
F. Daoust et al.,
“Clinical prototype Raman imaging system for intraoperative machine-learning-based molecular tissue margins characterization,”
Google Scholar
BiographySandryne David is a physics engineer and a doctoral student in biomedical engineering in the LRO Lab. Her research project involves the development of a Raman spectroscopy system allowing real-time breast cancer detection using machine learning to guide breast-conserving surgery. Her research interest is the development of new technologies based on spectroscopy to better health care. Trang Tran joined the LRO Lab as a research associate. With more than six years in research and specializing in neurosciences and biomedical with a focus on clinical applications, she currently oversees several clinical research projects in the laboratory including her main project researching the use of Raman spectroscopy on patients with focal cortical dysplasia. Her interests lie at the intersection of molecular biology and neurosurgery. Frédérick Dallaire is a physicist and a data scientist with a strong analysis and programming background. After his PhD in elementary particle physics, he joined the LRO Lab as a postdoc and has worked on the development of machine learning approaches using Raman spectroscopy for the identification and classification of brain tumors and prostate cancer. Guillaume Sheehy is a physics engineer and a doctoral student in biomedical engineering in the LRO Lab. His research project involves the development of numerical tools for better Raman spectra processing and for Raman spectra acquisition with complex spectroscopic systems as well as the creation of optical phantom and optical simulations. Feryel Azzi is a research pathologist, working in the molecular pathology laboratory of the University of Montreal Health Research Centre. Dominique Trudel is an associate clinical professor in the Faculty of Medicine, Department of Pathology and Cell Biology, University of Montreal. Her areas of expertise are cancers of the reproductive system, prostate cancer, and cancer of the urinary system. Francine Tremblay is an assistant professor of surgery and oncology at McGill University. Her clinical interests are surgical oncology, breast cancer, and melanoma. Atilla Omeroglu is the Director of Breast Pathology at McGill University Health Center and an assistant professor in the Department of Pathology, McGill University. Frédéric Leblond is a titular professor in the Department of Engineering Physics, where he heads the LRO Laboratory. His research interests are the development of medical instruments using light to characterize biological tissues to improve the accuracy and safety of surgical procedures, as well as to develop techniques to improve the accuracy of medical diagnoses. Sarkis Meterissian is a professor of surgery and oncology at McGill University and Director of the Breast Center of the McGill University Health Center. His research focuses on the diagnosis, pathogenesis, and prognosis of breast cancer and also to understand how a normal breast ductal epithelial cell becomes malignant. |
![Lens.org Logo](/images/Lens.org/lens-logo.png)
CITATIONS
Cited by 5 scholarly publications.
Cancer
Raman spectroscopy
Tissues
Breast
Machine learning
Tumor growth modeling
Surgery