A texture analysis technique improved classification of different types of breast tissue imaged with optical coherence microscopy (OCM), according to results of a study published in Medical Image Analysis.
OCM is a combination of optical coherence tomography and confocal microscopy. It provides improved penetration depth compared with confocal microscopy and better resolution than optical coherence tomography.
The imaging technique is emerging as a modality that could quicken tissue screening and provide resolution nearly as high as that in histology without the need to stain the tissue.
The researchers examined several types of texture features from 46 breast tissue samples; 27 were benign tissue (eg, fibroadenoma and usual ductal hyperplasia) and 19 were malignant (eg, invasive ductal carcinoma and carcinoma in situ).
The samples were viewed using large-field OCM. Corresponding histological analyses provided ground truth diagnoses.
In total, 4310 small OCM image blocks of 500 by 500 pixels, each paired with histology, were extracted from the large-field OCM images. Analyses labeled the smaller images as fat (n = 347), fibrous stroma (n = 2065), breast lobules (n = 199), carcinomas (n = 1127), and background (n = 572).
OCM can be used with a technique called local binary pattern (LBP). Here, researchers also determined accuracy of average LBP (ALBP) and block-based LBP (BLBP). Both imaging techniques demonstrated improved textural analyses.
By integrating ALBP and BLBP with LBP, the accuracy of classification improved from 81.7% with LBP only to 93.8% with a combination. Additional analyses indicated a sensitivity of 100% and specificity of 85.2% in differentiating benign from malignant tissue.
Wan S, Lee HC, Huang X, et al. Integrated local binary pattern texture features for classification of breast tissue imaged by optical coherence microscopy [published online March 8, 2017]. Med Image Anal. doi: 10.1016/j.media.2017.03.002