Quantitative characterization of human breast tissue based on deep learning segmentation of 3D optical coherence tomography images

Intraoperative Assessment of Breast Cancer Tissues after Breast-Conserving Surgery Based on Mapping the Attenuation Coefficients in 3D Cross-Polarization Optical Coherence Tomography

Intraoperative differentiation of tumorous from non-tumorous tissue can help in the assessment of resection margins in breast cancer and its response to therapy and, potentially, reduce the incidence of tumor recurrence. In this study, the calculation of the attenuation coefficient and its color-coded 2D distribution was performed for different breast cancer subtypes using spectral-domain CP OCT. A total of 68 freshly excised human breast specimens containing tumorous and surrounding non-tumorous tissues after BCS was studied. Immediately after obtaining structural 3D CP OCT images, en face color-coded attenuation coefficient maps were built in co-(Att(co)) and cross-(Att(cross)) polarization channels using a depth-resolved approach to calculating the values in each A-scan. We determined spatially localized signal attenuation in both channels and reported ranges of attenuation coefficients to five selected breast tissue regions (adipose tissue, non-tumorous fibrous connective tissue, hyalinized tumor stroma, low-density tumor cells in the fibrotic tumor stroma and high-density clusters of tumor cells). The Att(cross) coefficient exhibited a stronger gain contrast of studied tissues compared to the Att(co) coefficient (i.e., conventional attenuation coefficient) and, therefore, allowed improved differentiation of all breast tissue types. It has been shown that color-coded attenuation coefficient maps may be used to detect inter- and intra-tumor heterogeneity of various breast cancer subtypes as well as to assess the effectiveness of therapy. For the first time, the optimal threshold values of the attenuation coefficients to differentiate tumorous from non-tumorous breast tissues were determined. Diagnostic testing values for Att(cross) coefficient were higher for differentiation of tumor cell areas and tumor stroma from non-tumorous fibrous connective tissue: diagnostic accuracy was 91-99%, sensitivity-96-98%, and specificity-87-99%. Att(co) coefficient is more suitable for the differentiation of tumor cell areas from adipose tissue: diagnostic accuracy was 83%, sensitivity-84%, and specificity-84%. Therefore, the present study provides a new diagnostic approach to the differentiation of breast cancer tissue types based on the assessment of the attenuation coefficient from real-time CP OCT data and has the potential to be used for further rapid and accurate intraoperative assessment of the resection margins during BCS.

One-class machine learning classification of skin tissue based on manually scanned optical coherence tomography imaging

We investigated a method for automatic skin tissue characterization based on optical coherence tomography (OCT) imaging. We developed a manually scanned single fiber OCT instrument to perform in vivo skin imaging and tumor boundary assessment. The goal is to achieve more accurate tissue excision in Mohs micrographic surgery (MMS) and reduce the time required for MMS. The focus of this study was to develop a novel machine learning classification method to automatically identify abnormal skin tissues through one-class classification. We trained a deep convolutional neural network (CNN) with a U-Net architecture for automatic skin segmentation, used the pre-trained U-Net as a feature extractor, and trained one-class support vector machine (SVM) classifiers to detect abnormal tissues. The novelty of this study is the use of a neural network as a feature extractor and the use of a one-class SVM for abnormal tissue detection. Our approach eliminated the need to engineer the features for classification and eliminated the need to train the classifier with data obtained from abnormal tissues. To validate the effectiveness of the one-class classification method, we assessed the performance of our algorithm using computer synthesized data, and experimental data. We also performed a pilot study on a patient with skin cancer.

Digital histology of tissue with Mueller microscopy and FastDBSCAN

We present the results of the automated post-processing of Mueller microscopy images of skin tissue models with a new fast version of the algorithm of density-based spatial clustering of applications with noise (FastDBSCAN) and discuss the advantages of its implementation for digital histology of tissue. We demonstrate that using the FastDBSCAN algorithm, one can produce the diagnostic segmentation of high resolution images of tissue by several orders of magnitude faster and with high accuracy (>97%) compared to the original version of the algorithm.

Binary dose level classification of tumour microvascular response to radiotherapy using artificial intelligence analysis of optical coherence tomography images

The dominant consequence of irradiating biological systems is cellular damage, yet microvascular damage begins to assume an increasingly important role as the radiation dose levels increase. This is currently becoming more relevant in radiation medicine with its pivot towards higher-dose-per-fraction/fewer fractions treatment paradigm (e.g., stereotactic body radiotherapy (SBRT)). We have thus developed a 3D preclinical imaging platform based on speckle-variance optical coherence tomography (svOCT) for longitudinal monitoring of tumour microvascular radiation responses in vivo. Here we present an artificial intelligence (AI) approach to analyze the resultant microvascular data. In this initial study, we show that AI can successfully classify SBRT-relevant clinical radiation dose levels at multiple timepoints (t = 2–4 weeks) following irradiation (10 Gy and 30 Gy cohorts) based on induced changes in the detected microvascular networks. Practicality of the obtained results, challenges associated with modest number of animals, their successful mitigation via augmented data approaches, and advantages of using 3D deep learning methodologies, are discussed. Extension of this encouraging initial study to longitudinal AI-based time-series analysis for treatment outcome predictions at finer dose level gradations is envisioned.