Introduction

To treat breast cancer, diagnosis is necessary. Diagnostic biopsy relies on accuracy and classification thus the aim is to improve its procedural accuracy and surgical margin assessment. Missed diagnoses by false-negative biopsies range from 4.3% to 17.9%. The need for repeat biopsies occurs in 4% to 32% of patients. For breast conservative therapy, the need for a secondary surgical procedure is over 10% [1]. In order to increase sensitivity-specificity to discriminate between malignant and non-malignant tissue, breast tissue was analyzed using diffuse optical spectroscopy from a 500 to 1600 nm wavelength range to estimate its physiological, morphological, and optical parameters. The diffuse reflectance spectroscopy measurements and optical spectroscopy measured spectra from 102 ex vivo samples across the five tissue types of adipose, glandular, invasive carcinoma (IC), fibroadenoma (FA), and ductal carcinoma in situ (DCIS). The extracted parameters were blood, water, lipid, collagen volume fractions, β-carotene concentration. Optical spectroscopy uses light to determine and acquire the physical, chemical or structural properties of materials. It allows acquisition of information about a sample’s structure, organization, biochemistry, and physiology [2]. Diffuse optical spectroscopy examines the relationship between light and matter through concentration levels of tissues, and oxygen levels. It provides pathophysiological changes non-invasively while also being suitable for deep tissue measurement and portable [3].

Material and Methods

Breast tissue samples were collected from mastectomies and lumpectomies. Mastectomy is the removal of the breast. Mastectomy sample thickness ranged from 0.5 to 1 cm. Lumpectomy is the removal of a portion or “lump” of the breast. It is also known as breast conservative surgery. Lumpectomy or fibroadenoma (benign tumors) sample thickness were several mm in diameter. Table 1 summarizes the breakdown of breast tissue samples including number of samples and spectra.

Table 1. Histological description of breast tissue types and the corresponding amount of samples and spectra that were measured.

A large majority of breast tissue samples were non-malignant, the smallest samples and thereby spectra collected being fibroadenomas and ductal carcinomas in situ. Ex vivo diffuse reflectance spectra were taken using a portable spectroscopic system. A broadband light source with an integrated shutter delivered light into the tissue. Then, a 1.3 mm diameter fiber-optic probe with a distal end of 20 degrees collected the diffused light. This setup is seen below in Figure 1.

Fig. 1. Schematic of the optical setup and the design of the optical probe.

The probe was inserted into the sample. Wavelength values were assigned to each pixel of the detector. Integration time for each measurement was about 0.5 s. To model the measured spectra, diffusion theory was used to determine the absorption coefficients and reduced scattering coefficient. The reduced scattering coefficient is expressed as

where 0=800 nm, denotes the reduced scattering amplitude, b the Mie scattering slope, and the Mie-tot-total reduced scattering fraction. The absorption coefficient due to blood is expressed as

where Hb denotes the deoxygenated-hemoglobin, HbO2 the oxygenated-hemoglobin, v the blood volume fraction, StO2 the level of hemoglobin saturation by oxygen, and C() the inhomogeneous distribution of hemoglobin or pigment packaging factor. Accordingly, he pigment packaging factor is expressed as

where R is the average vessel radius. The absorption coefficient due to non-blood derived chromophores is

where denotes the water and lipid volume fraction, f the fraction within volume or tissue, and cc the molar concentration of -carotene. Using these equations, the absorption coefficients can be seen in Figure 2.

Fig. 2. Normalized absorption coefficients of Hb, HbO2, β-carotene, water (H2O), lipid, and collagen.

Results

The CART or Classification and Regression Trees algorithm was used to classify between the five types of tissue (adipose, glandular, fibroadenoma, invasive carcinoma, and ductal carcinoma). The tree starts from a central node that discriminates the largest class based on the best classifier. From this root node, it splits to discriminate the largest class from the other tissue classes and further splits into daughter partial trees using other parameters. For the study the largest class is adipose tissue. The CART classifies all tissues based on a specific threshold value for each parameter as seen in Figure 4.

Fig. 4. Classification decision tree of the different breast types based on parameter threshold values.

A confusion matrix shows diagnostic performance by comparing with the pathological diagnosis being the reference standard. It utilizes machine learning to create a prediction summary. The confusion matrix for the study is seen in Table 2.

Table 2. Confusion matrix displaying classification of breast tissues using the CART algorithm for classification.

As seen in the bolded numbers, the predictions showed that fibroadenoma tissue had the lowest sensitivity rate while adipose tissue had the highest specificity rate. The plot of sensitivity versus specificity is called the receiver operating characteristic (ROC) curve and the area under the curve (AUC). It provides a measure of accuracy that can be interpreted to “discriminate the true state of subjects, finding the optimal cut off values, and comparing two alternative diagnostic tasks when each task is performed on the same subject” [4]. The corresponding AUC of ROC of the different tissue types is seen in Figure 5.

Fig. 5. ROC curves (solid line) for classification of adipose (a), glandular (b), FA (c), IC (d), and DCIS (e) tissues including confidence intervals (dashed line) and corresponding AUC

The effectiveness or performance of the diagnoses can be classified using the AUC values. As seen in the figure, adipose tissue had the best performance with AUC of almost 100%. On the other hand, glandular tissue had the worst performance with AUC of 86%. The other tissue (fibroadenoma, invasive carcinoma, and ductal carcinoma in situ) had median performance with comparable AUC values around 92%. Lipids were identified as a better discriminator for adipose tissue with sensitivity–specificity of 98%–99% versus 68%–92% for β-carotene. Based on the CART method used in Figure 4, discriminating adipose tissue based on the β-carotene values only yielded a sensitivity-specificity of 68%–92%, whereas classification based on the lipid parameters yielded to a sensitivity-specificity of 98%–99%. However, using both parameters yielded the best results with a sensitivity-specificity of 99%–99%. The combination of methods helped improve diagnostic accuracy or performance.

References

[1] Rami Nachabe et. al, “Diagnosis of breast cancer using diffuse optical spectroscopy from 500 to 1600 nm: comparison of classification methods,” in Journal of Biomedical Optics, Vol. 16(8), pp 01-11.

[2] Irene Georgakoudi et. al, “Optical Spectroscopy and Imaging for the Noninvasive Evaluation of Engineered Tissues” in Tissue Eng Part B Rev, Vol 14(4), pp. 321-340.

[3] So Hyun Chung, “Diffuse Optical Technology: A Portable and Simple Method for Noninvasive Tissue Pathophysiology,” in PET Clin.

[4] Karimollah Hajian-Tilaki, “Receiver Operating Characteristic (ROC) Curve Analysis for Medical Diagnostic Test Evaluation,” in Caspian J Intern Med, Vol 4(2), pp. 627-635.