Histogram analysis of en face scattering coefficient map predicts malignancy in human ovarian tissue

Abstract

Ovarian cancer is a heterogeneous disease at the molecular and histologic level. Optical coherence tomography (OCT) is able to map ovarian tissue optical properties and heterogeneity, which has been proposed as a feature to aid in diagnosis of ovarian cancer. In this manuscript, depth-resolved en face scattering maps of malignant ovaries, benign ovaries, and benign fallopian tubes obtained from 20 patients are provided to visualize the heterogeneity of ovarian tissues. Six features are extracted from histograms of scattering maps. All features are able to statistically distinguish benign from malignant ovaries. Two prediction models were constructed based on these features: a logistic regression model (LR) and a support vector machine (SVM). The optimal set of features is mean scattering coefficient and scattering map entropy. The LR achieved a sensitivity and specificity of 97.0% and 97.8%, and SVM demonstrated a sensitivity and specificity of 99.6% and 96.4%. Our initial results demonstrate the feasibility of using OCT as an "optical biopsy tool" for detecting the microscopic scattering changes associated with neoplasia in human ovarian tissue.

Keywords: cancer prediction; optical coherence tomography; ovarian cancer; scattering coefficient map.

FIGURE 1

Representative OCT images of benign and malignant ovary specimens. The ovarian surface epithelium is between the two red curves in the B-scan SS-OCT images. (A) representative B-scan image from a benign ovary and (B) corresponding H&E image. (C) representative B-scan image from a malignant ovary (high-grade serous carcinoma) and (D) corresponding H&E image. Inset: best-fit Beer’s law is used to calculate the scattering coefficient. OCT, optical coherence tomography

FIGURE 2

Photographs (A-C) of one benign ovary, one malignant ovary, and one benign fallopian tube, respectively. Scattering coefficient maps (D-F) of the scanned areas, identified as white boxes in Figure (A-C). The scale bar of 200 mm is shared by maps (D-F)

FIGURE 3

Histogram analysis of one representative malignant ovary (A) and one representative benign ovary (B). The six features for (A) are 4.0 mm⁻¹(mean), 1.71(variance), 6.50(entropy), 0.77(skewness), 4.33(kurtosis), and 0.17(energy). The six features for (B) are 11.48mm⁻¹(mean), 2.98(variance), 7.30(entropy), 0.22(skewness), 2.85(kurtosis), and 0.10(energy). Fitted Gaussian distribution is shown as red curves

FIGURE 4

Boxplot of the six features extracted from histogram analysis of scattering maps of malignant ovaries, benign ovaries, and benign fallopian tubes

FIGURE 5

Testing results of two optimal data sets ([mean, entropy] and [energy, skewness, entropy]) used to train two classification models. (A-B) show the ROC curves for the testing sets of logistic regression and (C-D) show the ROC curves of SVM model