Detection of breast cancer using Raman spectroscopy

Introduction and Motivation
Breast cancer accounts for nearly one in every three cancers diagnosed in American women and, excluding cancers of the skin, breast is the most common cancer among women. Mammography is the most common technique for detecting non-palpable, highly curable breast cancer. Mammography quantitatively probes density changes in breast tissue, however, theses changes are not uniquely correlated with the probability of breast cancer. Because of this, it serves as a screening technique rather than as a diagnostic tool. This is evidenced by the fact that only 10-25% of mammographically detected lesions are found to be malignant upon needle biopsy. As a consequence of the limitations of current techniques, each year a large number of breast biopsies are performed on benign lesions. The desirability of reducing the number of benign biopsies preformed, patient trauma, time delay, and the high medical costs associated with biopsy has motivated researchers to explore minimally invasive optical methods for diagnosing malignant lesions in the breast.

Raman scattering is a spectroscopic technique capable of providing highly detailed chemical information about a tissue sample. In contrast to other optical spectroscopic techniques, there are a large number of Raman active molecules in breast tissue and their spectral signatures are sharp and well delineated. The ability to probe several different chemicals is of particular importance in studying breast cancer, due to the heterogeneity of breast disease. Further, Raman spectroscopy is particularly amenable to in vivo measurements as the powers and excitation wavelengths used are non-destructive to the tissue and have a relatively large penetration depth. For these reasons, we have investigated the use of Raman spectroscopy as a clinical tool for the examination of a variety of breast pathologies.

Morphological Raman Spectral Model of Breast Tissue
In an effort to understand the relationship between a tissue sample’s Raman spectrum and its disease state we have developed a chemical/morphological model of breast tissue. This model fits macroscopic tissue spectra, sampling volume 1 mm3, with a linear combination of basis spectra derived from Raman microscopy of various breast tissue morphological structures. The structures studied are the cell cytoplasm, the cell nucleus, fat, beta-carotene, collagen, calcium hydroxyapatite, calcium oxalate dihydrate, and cholesterol-like lipid deposits. Their Raman spectra are shown in Figure 1. To obtain the basis spectra we collected Raman maps from thin sections (8?m) of normal and diseased human breast tissues. These maps were then compared with phase contrast and hematoxylin- and eosin-stained images to correlate Raman spectra and morphological features. Data were acquired with a confocal Raman microspectrometer, sampling volume (2 ?m)3. This approach allows for the creation of a library of Raman spectra from individual morphological structures and determination of their chemical composition. Each basis spectrum in the model represents data acquired from multiple patients and, when appropriate, from a variety of normal and diseased states.

Figure 1. Raman morphological model basis spectra.

Modeling is based on the assumptions that the Raman spectrum of a mixture is a linear combination of the spectra of its components and that signal intensity and chemical concentration are linearly related. Ordinary least-squares fitting of the macroscopic Raman spectrum of tissues yields the contribution of each basis spectrum to the entire tissue spectrum, thereby elucidating the chemical/morphological makeup of the lesion. These same morphological changes are routinely used by pathologists to diagnose disease. However, unlike conventional pathology, Raman spectroscopy assesses these changes in an objective and reproducible manner. Our model explains the spectral features of a range of normal and diseased breast tissue samples and can be used to relate the Raman spectrum of a breast tissue sample to diagnostic parameters used by pathologists.

Diagnosing Breast Cancer Using Raman Spectroscopy
Our spectral model was used to characterize the chemical/morphological composition of different breast pathologies. We analyzed Raman spectra from a range of ex vivo tissue samples and pathologies to assess the ability of the model to predict breast tissue disease state. In this study, a total of 130 Raman spectra were acquired from intact, ex vivo samples of human breast. The volume sampled was 1 mm3. The Raman spectra were fit to a linear combination of the basis spectra using an ordinary least-squares minimization algorithm with a non-negativity constraint. The fit coefficients given by the model and normalized to sum to one, represent contributions of chemicals and morphological features to the macroscopic tissue spectrum.

Figure 2 displays a histogram of the average fit coefficients associated with normal breast tissue, fibrocystic change, fibroadenoma, and infiltrating carcinoma. The fit coefficients of normal breast tissue indicate that it is predominantly composed of fat. This is because normal breast tissue is largely made up of adipocytes, cells containing copious amounts of cytoplasmic fat, and adipose tissue has a relatively large Raman scattering cross-section. The fit coefficients of the breast lesions indicate a markedly different chemical/morphological composition than that of normal breast tissue. First, the amount of collagen increases in all abnormal breast pathologies. This is consistent with known breast pathology, as lesion formation is often accompanied by fibrosis. Fibrosis is a scarring process characterized by an increased stromal component, and thus an accumulation of collagen. The relative increase in collagen is most pronounced in fibrocystic change, a benign condition that can manifest as fibrosis, adenosis (increase in the number of ductules) and/or cyst formation. Fibroadenoma is a benign tumor characterized by both fibroblast and ductal proliferation.

We observe an increased contribution from both the cell nucleus and epithelial cell cytoplasm basis spectra in samples diagnosed as fibroadenoma as a consequence of the number of fibroblasts and epithelial cells present. The fit coefficients of lesions diagnosed as infiltrating carcinoma, also a proliferative lesion, indicate an increase in the amount of epithelial cell cytoplasm and cell nucleus. Differences between lesions diagnosed as fibroadenoma and infiltrating carcinoma exist in the amount of fat present. Samples diagnosed as fibroadenoma have less fat than those diagnosed as infiltrating carcinoma, as can be seen in Figure 2. This is because fibroadenoma is an expansile lesion which grows by pushing fatty breast tissue aside. Infiltrating carcinoma, on the other hand, infiltrates in between the fat cells, so some adipocytes will be retained within the carcinoma.

The fit coefficients of the basis spectra not only provide insight into the chemical/morphological makeup of the tissue but were also used to develop diagnostic algorithms. Logistic regression, a discriminate analysis technique, was used to correlate the normalized fit coefficients with the diagnostic categories. A leave-one-out cross validation analysis was employed in order to obtain a robust diagnostic algorithm. Following this strategy, we are able to distinguish normal and malignant tissues with a sensitivity 100% and a specificity of 100% based on their fat and collagen content. Benign diagnoses are grouped into two categories, fibroadenoma and fibrocystic change, the two most common benign breast pathologies. Malignant lesions are separated from fibrocystic change and fibroadenoma with sensitivities of 94% and 90% and specificities of 94% and 80%, respectively.

A single algorithm encompassing all data is shown in Figure 3. It classifies samples according to specific pathological diagnoses and was used to identify those lesions that would need to be excised. This algorithm resulted in a negative predictive value of 100%, indicating that no malignant lesions were left unexcised in our data set. Using this algorithm, 10 out of every 100 benign lesions examined would be unnecessarily removed. The excellent results of this proof of principle study support moving the technique to a clinical setting for further testing of its efficacy in breast cancer detection. Planning for clinical studies is currently underway.

Recent Publications

• A.S. Haka, K.E. Shafer-Peltier, M. Fitzmaurice, J. Crowe, R.R. Dasari, and M.S. Feld, “Detecting Breast Cancer Using Raman Spectroscopy”, P. Natl. Acad. Sci. USA, submitted (2004).
• A.S. Haka, K.E. Shafer-Peltier, M. Fitzmaurice, J. Crowe, R.R. Dasari, and M.S. Feld, “Identifying Microcalcifications in Benign and Malignant Breast Lesions by Probing Differences in their Chemical Composition Using Raman Spectroscopy” Cancer Research, 62 18 (2002).
• K.E. Shafer-Peltier, A.S. Haka, M. Fitzmaurice, J. Crowe, J. Myles, R.R. Dasari, and M.S. Feld, “Raman Microspectroscopic Model of Human Breast Tissue: Implications for Breast Cancer Diagnosis in vivo” J. Raman Spec. 33 (2002).
• K.E. Shafer-Peltier, A.S. Haka, J.T. Motz, M. Fitzmaurice, R.R. Dasari, and M.S. Feld, “Model-based Biological Raman Spectral Imaging”, J. of Cellular Biochem. 87 S125-S137 (2002).
• E.B. Hanlon, R. Manoharan, T.W. Koo, K.E. Shafer, J.T. Motz, M. Fitzmaurice, J.R. Kramer, I. Itzkan, R.R. Dasari, and M.S. Feld, "Prospects of in vivo Raman spectroscopy", Phys. Med. Biol. 45: R1-R59 (2000).
• R. Manoharan, K.E. Shafer, L. Perelman, J. Wu, K. Chen, G. Deinum, M. Fitzmaurice, J. Myles, J. Crowe, R.R. Dasari, and M.S. Feld, "Raman spectroscopy and fluorescence photon migration for breast cancer diagnosis and imaging", Photochem. Photobiol. 67 1, 15-22 (1998).