Integrated local binary pattern texture features for classification of breast tissue imaged by optical coherence microscopy

Abstract

This paper proposes a texture analysis technique that can effectively classify different types of human breast tissue imaged by Optical Coherence Microscopy (OCM). OCM is an emerging imaging modality for rapid tissue screening and has the potential to provide high resolution microscopic images that approach those of histology. OCM images, acquired without tissue staining, however, pose unique challenges to image analysis and pattern classification. We examined multiple types of texture features and found Local Binary Pattern (LBP) features to perform better in classifying tissues imaged by OCM. In order to improve classification accuracy, we propose novel variants of LBP features, namely average LBP (ALBP) and block based LBP (BLBP). Compared with the classic LBP feature, ALBP and BLBP features provide an enhanced encoding of the texture structure in a local neighborhood by looking at intensity differences among neighboring pixels and among certain blocks of pixels in the neighborhood. Fourty-six freshly excised human breast tissue samples, including 27 benign (e.g. fibroadenoma, fibrocystic disease and usual ductal hyperplasia) and 19 breast carcinoma (e.g. invasive ductal carcinoma, ductal carcinoma in situ and lobular carcinoma in situ) were imaged with large field OCM with an imaging area of 10 × 10 mm² (10, 000 × 10, 000 pixels) for each sample. Corresponding H&E histology was obtained for each sample and used to provide ground truth diagnosis. 4310 small OCM image blocks (500 × 500 pixels) each paired with corresponding H&E histology was extracted from large-field OCM images and labeled with one of the five different classes: adipose tissue (n = 347), fibrous stroma (n = 2,065), breast lobules (n = 199), carcinomas (pooled from all sub-types, n = 1,127), and background (regions outside of the specimens, n = 572). Our experiments show that by integrating a selected set of LBP and the two new variant (ALBP and BLBP) features at multiple scales, the classification accuracy increased from 81.7% (using LBP features alone) to 93.8% using a neural network classifier. The integrated feature was also used to classify large-field OCM images for tumor detection. A receiver operating characteristic (ROC) curve was obtained with an area under the curve value of 0.959. A sensitivity level of 100% and specificity level of 85.2% was achieved to differentiate benign from malignant samples. Several other experiments also demonstrate the complementary nature of LBP and the two variants (ALBP and BLBP features) and the significance of integrating these texture features for classification. Using features from multiple scales and performing feature selection are also effective mechanisms to improve accuracy while maintaining computational efficiency.

Keywords: Local binary patterns; Optical coherence microscopy; Texture features; Tissue classification.

Figure 1

Sample OCM images (first row) and corresponding histology images (second row) of human breast tissue. The ground truth labels for the tissue type of these images are: (a) carcinoma, (b) lobule, (c) stroma, (d) adipose.

Figure 2

Overview diagram for the training and testing processes.

Figure 3

Demonstration of LBP feature in a local neighborhood of an OCM image.

Figure 4

Two OCM images with the same LBP feature but different ALBP features. (a),(c) are image 1; (b),(d) are image 2; (a),(b) show LBP features; (c),(d) show ALBP features.

Figure 5

Integrated LBP and ALBP feature vectors for OCM images of two types of tissue. In (c), columns 1–9 represent LBP feature and columns 10–18 represent ALBP feature.

Figure 6

Demonstrated of Block based LBP (BLBP) feature.

Figure 7

BLBP feature vectors of the lubule and fat tissue shown in 5.

Figure 8

Integrated multi-scale LBP+ALBP+BLBP feature vector: columns 1–9: ${\mathit{LBP}}_{8,16}^{\mathit{riu}2}$, columns 10–18: ${\mathit{LBP}}_{8,8}^{\mathit{riu}2}$, columns 19–27: ${\mathit{LBP}}_{8,4}^{\mathit{riu}2}$, columns 28–36: ${\mathit{LBP}}_{8,2}^{\mathit{riu}2}$, columns 37–45: ${\mathit{ALBP}}_{8,16}^{\mathit{riu}2}$, columns 46–54: ${\mathit{ALBP}}_{8,8}^{\mathit{riu}2}$, columns 55–63: ${\mathit{ALBP}}_{8,4}^{\mathit{riu}2}$, columns 64–72: ${\mathit{ALBP}}_{8,2}^{\mathit{riu}2}$, columns 73–81: SBLBP_8,3,, columns 82–90: SBLBP_8,6, columns 91–99: SBLBP_8,12, columns 100–108: SBLBP_8,18, columns 109–116: RBLBP_8,1,3, columns 117–124: RBLBP_8,2,3, columns 125–132: RBLBP_8,4,3, columns 133–140: RBLBP_8,6,3.

Figure 9

Examples from UIUCTex texture image database.

Figure 10

Examples from CUReT texture image database.

Figure 11

ROC curve for tumor tissue detection in large-field OCM images.

Figure 12

Breast tissue OCM image classification results. The results include two classes of images: image of tissue with tumor (image 1 and 3), and image of tissue without tumor (image 2). (a) histology image; (b) OCM image; (c) classification result; (d) probability distribution (i.e. heat map) of tumor tissue.

Figure 13

Average feature values for five different classes of OCM image blocks and demonstration of attribute selection. The total dimension of a feature vector is 140. 24 attributes are selected by the information gain based attribute selection algorithm. The selected 24 are marked by red squares along the horizontal axis. Dimension 1–9: ${\mathit{LBP}}_{8,16}^{\mathit{riu}2}$, dimension 10–18: ${\mathit{LBP}}_{8,8}^{\mathit{riu}2}$, dimension 19–27: ${\mathit{LBP}}_{8,4}^{\mathit{riu}2}$, dimension 28–36: ${\mathit{LBP}}_{8,2}^{\mathit{riu}2}$, dimension 37–45: ${\mathit{ALBP}}_{8,16}^{\mathit{riu}2}$, dimension 46–54: ${\mathit{ALBP}}_{8,8}^{\mathit{riu}2}$, dimension 55–63: ${\mathit{ALBP}}_{8,4}^{\mathit{riu}2}$, dimension 64–72: ${\mathit{ALBP}}_{8,2}^{\mathit{riu}2}$, dimension 73–81:SBLBP_8,3, dimension 82–90:SBLBP_8,6, dimension 91–99:SBLBP_8,12, dimension 100–108:SBLBP_8,18, dimension 109–116:RBLBP_8,1,3, dimension 117–124:RBLBP_8,2,3, dimension 125–132:RBLBP_8,4,3, dimension 133–140:RBLBP_8,6,3.