Optimizing the classification of biological tissues using machine learning models based on polarized data

Polarimetric data is nowadays used to build recognition models for the characterization of organic tissues or the early detection of some diseases. Different Mueller matrix-derived polarimetric observables, which allow a physical interpretation of a specific characteristic of samples, are proposed in literature to feed the required recognition algorithms. However, they are obtained through mathematical transformations of the Mueller matrix and this process may loss relevant sample information in search of physical interpretation. In this work, we present a thorough comparative between 12 classification models based on different polarimetric datasets to find the ideal polarimetric framework to construct tissues classification models. The study is conducted on the experimental Mueller matrices images measured on different tissues: muscle, tendon, myotendinous junction and bone; from a collection of 165 ex-vivo chicken thighs. Three polarimetric datasets are analyzed: (A) a selection of most representative metrics presented in literature; (B) Mueller matrix elements; and (C) the combination of (A) and (B) sets. Results highlight the importance of using raw Mueller matrix elements for the design of classification models.     

Investigation on the potential of Mueller matrix imaging for digital staining

Digital staining based on Mueller matrix measurements and their derivatives was investigated. Mueller matrix imaging was performed at the microscopic level on gastric tissue sections. Full Mueller matrices (4 × 4) were reconstructed using recorded images, followed by the extraction of polarization parameters. The most effective parameters and their combinations were extracted from Mueller matrix elements, principal component scores and polarization parameters respectively to classify samples into three categories – i.e. cancer, dysplasia and intestinal metaplasia/normal glands for various regions of interest sizes. It was observed that two‐step classification yielded higher classification accuracy than the traditional one‐step classification and that pixel classification based on Mueller matrix elements yielded higher accuracy than that based on polarization parameters and derived principal components. Moreover, Mueller matrix images with a lower spatial resolution generated higher classification accuracy but those with a higher spatial resolution revealed more morphological details.ns.

The original stained image (top) and the digital staining image (bottom).     

Polarimetric optical scanning microscopy of zebrafish embryonic development using the coherency matrix

Many of the most important resolution improvements in optical microscopy techniques are based on the reduction of scattering effects. The main benefit of polarimetry-based imaging to this end is the discrimination between scattering phenomena originating from complex systems and the experimental noise. The determination of the coherency matrix elements from the experimental Mueller matrix can take advantage of scattering measurements to obtain additional information on the structural organization of a sample. We analyze the contrast mechanisms extracted from (a) the coherency matrix elements, (b) its eigenvalues and (c) the indices of polarimetric purity at different stages of zebrafish embryos, based on previous work using Mueller matrix optical scanning microscopy. We show that the use of the coherency matrix and related decompositions leads to an improvement in the imaging contrast, without requiring any complicated algebraic operations or any a priori knowledge of the sample, in contrast to standard polarimetric methods.     

Mapping of retardance,diattenuation and polarizance vector on Poincare sphere for diagnosis and classification of cervical precancer

The mapping of diattenuation, polarizance and retardance vector (normalized Stokes vector) on Poincare sphere, evaluated from Mueller matrix of optically anisotropic stromal region of cervical tissues, is presented for cervical precancer detection and its staging. This reveals that the changes in the polarization states shown by these normalized Stokes vectors corresponds to the degradation of linearly arranged collagen fibers, breakage of the collagen cross links in the stromal region and change in the density of scattering sites when cervical cancer evolves. The distinct nature of real and imaginary parts of the refractive index for linear, linear‐45 and circular polarization from the optically anisotropic stromal region underscore the various polarization structures of the connective tissue region which get modified during the pathological changes. It has been found that versatility of these vectors for normal and precancerous cervical tissue of various grades may be utilized as a key distinction for qualitative staging of cervical precancer tissue. Quantitative classification of precancerous stages of cervical precancer has been determined with 95%–100% sensitivity and 93%–100% specificity through the evaluation of linear and circular diattenuation, linear polarizance and linear birefringence from the components of the respective vectors.     

Distinguishing structural features between Crohn's disease and gastrointestinal luminal tuberculosis using Mueller matrix derived parameters

Recently, the incidence of inflammatory bowel diseases, especially the Crohn's disease (CD) and gastrointestinal luminal tuberculosis (ITB), has grown rapidly worldwide. Currently there is no general gold standard to distinguish between CD and ITB tissues, which both have tuberculosis and surrounding fibrous structures. Mueller matrix imaging technique is suitable for describing the location, density and distribution behavior of such fibrous structures. In this study, we apply the Mueller matrix microscopic imaging to the CD and ITB tissue samples. The 2D Mueller matrix images of the CD and ITB tissue slices are measured using the Mueller matrix microscope developed in our previous study, then the Mueller matrix polar decomposition and Mueller matrix transformation parameters are calculated. To evaluate the distribution features of the fibrous structures surrounding the tuberculosis areas more quantitatively and precisely, we analyze the retardance related Mueller matrix derived parameters, which show clear different distribution behaviors between the CD and ITB tissues, using the Tamura image processing method. It is demonstrated that the Mueller matrix derived parameters can reveal the structural features of tuberculosis areas and be used as quantitative indicators to distinguish between CD and ITB tissues, which may be useful for the clinical diagnosis.      

Evolution of raw meat polarization-based properties by means of Mueller matrix imaging

The possibilities of using Mueller matrix (MM) imaging polarimetry to assess meat quality have not yet been sufficiently explored. In the current study, the fresh porcine muscles are imaged at room temperature with a wide-field MM imaging polarimeter over 26 hours to visualize dynamics of tissue optical properties through applying Lu-Chipman decomposition. The frequency distribution histograms (FDHs) and statistical analysis of the MM elements show prominent changes over time. The wavelength spectra of both total depolarization and scalar retardance have dips at 550 nm whereas their values continuously increase with time; the former is referred to the increase of number of scattering events and decrease of myoglobin absorption in the red part of visible spectra related to meat color and freshness, while the latter is associated with the increase in birefringence and meat tenderness. The obtained results are promising to develop a novel fast noncontact optical technique for monitoring of meat quality.     

Spatial autocorrelation analysis on two‐dimensional images of Mueller matrix for diagnosis and differentiation of cervical precancer

The spatial autocorrelation and correlation map of amplitude and phase anisotropy along with depolarization parameter from the stroma of uterine cervix utilizing their Mueller matrix (MM) images have been reported for early diagnosis of cervical cancer and differentiation of precancerous stages. The comparative results of the evaluation of the spatial autocorrelation over MM images of optically anisotropic collagen structures from normal and various grades of cervical precancer reflect significant alterations which are correlated with the pathological changes. The spatially varying polarizance from different region of anisotropic stromal region gets correlated within a given spatial lag during the precancerous changes. The diattenuation governing elements M12, M13 and M14 clearly discriminate normal and various grades of precancerous cervical tissue through their autocorrelation profile and correlation map. Evaluation of autocorrelation of spatially varying linear birefringence and linear‐45 birefringence characterized by MM elements M34 and M43 and M24 and M42 are not found to differ between the precancer grades, indicating that these changes may be arising from highly directional collagen network while the changes displayed by MM elements M23 and M32 faithfully represent that the chirality of the stromal region is compromised as the cervical cancer evolves and only one type of nature dominates.     

Synchrotron Fourier-Transform Infrared Microspectroscopy: Characterization of in vitro polarized tumor-associated macrophages stimulated by the secretome of inflammatory and non-inflammatory breast cancer cells

Studies suggested that the pathogenesis of inflammatory breast cancer (IBC) is related to inflammatory manifestations accompanied by specific cellular and molecular mechanisms in the IBC tumor microenvironment (TME). IBC is characterized by significantly higher infiltration of tumor-associated macrophages (TAMs) that contribute to its metastatic process via secreting many cytokines such as TNF, IL-6, IL-8, and IL-10 that enhance invasion and angiogenesis. Thus, there is a need to first understand how IBC-TME modulates the polarization of TAMs to better understand the role of TAMs in IBC. Herein, we used gene expression signature and Synchrotron Fourier-Transform Infrared Microspectroscopy (SR-μFTIR) to study the molecular and biochemical changes, respectively of in vitro polarized TAMs stimulated by the secretome of IBC and non-IBC cells. The gene expression signature showed significant differences in the macrophage's polarization-related genes between stimulated TAMs. FTIR spectra showed absorption bands in the region of 1700–1500 cm^?1 attributed to the amide I ν(C=O), & ν_AS (CN), δ (NH), and amide II ν(CN), δ (NH) proteins bands. Moreover, three peaks of different intensities and areas were detected in the lipid region of the νCH₂ and νCH₃ stretching modes positioned within the 3000–2800 cm^?1 range. The PCA analysis for the second derivative spectra of the amide regions discriminates between stimulated IBC and non-IBC TAMs. This study showed that IBC and non-IBC TMEs differentially modulate the polarization of TAMs and SR-μFTIR can determine these biochemical changes which will help to better understand the potential role of TAMs in IBC.     

Monitoring microstructural variations of fresh skeletal muscle tissues by Mueller matrix imaging

Recently many attempts have been made for extracting the structural information of myofibrils as indicators for diseases of skeletal muscle. In this paper we adopt wide‐field illumination and take the backscattering Mueller matrix images of bovine skeletal muscle tissues during the 24‐hour experimental time after the animal's death. The 2D images of Mueller matrix elements and their frequency distribution histograms (FDHs) reveal rich qualitative information on the changes in the microstructures of the skeletal muscle. The temporal variations of the sample are quantitatively analyzed using two Mueller matrix transformation (MMT) parameters. The characteristic features of the temporal plots are attributed to the rigor mortis and proteolysis processes. For a deeper insight on the relationship between the features of the MMT parameters and the microstructures during the rigor mortis and proteolysis processes, Monte Carlo (MC) simulations are carried out based on sphere‐cylinder birefringence model (SCBM). The good agreement between the experimental and MC simulated results show that the FDHs and MMT parameters can describe more clearly the characteristic microstructural features of skeletal muscle tissues. The techniques are useful for the characterization of physiological status of tissues, or quantitative assessment of meat qualities in food industry.

     

Automated thermal imaging monitors the local response to cervical cancer brachytherapy

Malignant tumors have high metabolic and perfusion rates, which result in a unique temperature distribution as compared to healthy tissues. Here, we sought to characterize the thermal response of the cervix following brachytherapy in women with advanced cervical carcinoma. Six patients underwent imaging with a thermal camera before a brachytherapy treatment session and after a 7-day follow-up period. A designated algorithm was used to calculate and store the texture parameters of the examined tissues across all time points. We used supervised machine learning classification methods (K Nearest Neighbors and Support Vector Machine) and unsupervised machine learning classification (K-means). Our algorithms demonstrated a 100% detection rate for physiological changes in cervical tumors before and after brachytherapy. Thus, we showed that thermal imaging combined with advanced feature extraction could potentially be used to detect tissue-specific changes in the cervix in response to local brachytherapy for cervical cancer.     

Front Cover: Differentiating smokers and nonsmokers based on Raman spectroscopy of oral fluid and advanced statistics for forensic applications (J. Biophotonics 3/2020)

Raman spectroscopy was employed as a nondestructive technique for the analysis of dry traces of oral fluid to differentiate between smoker and nonsmoker donors with the aid of advanced statistical tools. A genetic algorithm (GA) was used to determine eight spectral regions that contribute the most to the differentiation of smokers and nonsmokers. Thereafter, a classification model was developed based on the artificial neural network (ANN) that showed 100% accuracy after external validation. (Image ©: Dr. Aliaksandra Sikirzhytskaya). Further details can be found in the article by Entesar Al‐Hetlani, Lenka Halamkova, Mohamed O. Amin, and Igor K. Lednev ( e201960123 ).

     

EZH2 and ALDH1 expression in ductal carcinoma in situ: Complex association with recurrence and progression to invasive breast cancer

The diagnosis of ductal carcinoma in situ (DCIS) is an increasingly common event due to widespread use of screening mammography. However, appropriate clinical management of DCIS is a major challenge in the absence of prognostic markers. Tumor-initiating cells may be particularly relevant for disease pathogenesis; therefore, two markers associated with such cells, EZH2 and ALDH1, were evaluated. A cohort of 248 DCIS patients was used to determine the association of EZH2 and ALDH1 with ipsilateral breast event, DCIS recurrence and progression to invasive breast cancer (IBC). In this cohort, high EZH2 expression was associated with the risk of an ipsilateral breast event and DCIS recurrence but not invasive progression. ALDH1 expression was observed in both the tumor and stromal compartment; however, in neither compartment were ALDH1 levels independently associated with evaluated study endpoints. Interestingly, the combination of high EZH2 with high epithelial ALDH1 was associated with disease progression. Therefore, ALDH1 within the DCIS lesion can add to the prognostic significance of EZH2, particularly in the context of risk of development of invasive disease.     

Integrated photoacoustic and hyperspectral dual‐modality microscopy for co‐imaging of melanoma and cutaneous squamous cell carcinoma in vivo

Skin carcinoma such as melanoma (MM) and cutaneous squamous cell carcinoma (cSCC) are considered as the highest mortality and the most aggressive skin cancers in dermatology. In view that early diagnosis and treatment can greatly improve the survival rate and life quality of the patients, developing noninvasive and effective evaluation methods is of great significance for the detection and identification of early stage cutaneous cancers. In this article, we propose a hybrid photoacoustic and hyperspectral dual‐modality microscopy to evaluate and differentiate skin carcinoma by structural and multiphysiological parameters. The proposed system's imaging abilities are verified by mimic phantoms and normal mice experiments. Furthermore, in vivo characterization and evaluation results of MM and cSCC mice are obtained successfully, which prove this novel method could be used as a reliable and useful method for skin cancer detection in early stages.     

3D-PulCNN: Pulmonary cancer classification from hyperspectral images using convolution combination unit based CNN

Pulmonary cancer is one of the most common malignancies worldwide. Accurate classification of its subtypes is required in differential diagnosis. However, existing algorithms are mostly based on color images, and the improvement of accuracy is quite challenging. In this study, we propose a convolution combination unit (CCU)-based three-dimensional convolutional neural network (3D-PulCNN) for classifying pulmonary cancer presented in microscopic hyperspectral image with both spatial and spectral information. CCU is designed to fuse the features acquired by different convolution scales. Compared with VGGNet, only two fully connected layers are used in this model, reducing the network parameters and model complexity. Experimental results show that 3D-PulCNN achieves overall average (OA) of 0.962 and Precision, Recall, and Kappa of more than 0.920, superior to 2D-VGGNet. Then, 3D-UNet is leveraged to segment cancer cells, and their morphological characteristics are calculated to supply quantitative virtual analysis data for classification results explanation and prognosis assessment.     

Constitutive activation of smoothened (SMO) in mammary glands of transgenic mice leads to increased proliferation, altered differentiation and ductal dysplasia

The hedgehog signaling network regulates pattern formation, proliferation, cell fate and stem/progenitor cell self-renewal in many organs. Altered hedgehog signaling is implicated in 20-25% of all cancers, including breast cancer. We demonstrated previously that heterozygous disruption of the gene encoding the patched-1 (PTCH1) hedgehog receptor, a negative regulator of smoothened (Smo) in the absence of ligand, led to mammary ductal dysplasia in virgin mice. We now show that expression of activated human SMO (SmoM2) under the mouse mammary tumor virus (MMTV) promoter in transgenic mice leads to increased proliferation, altered differentiation, and ductal dysplasias distinct from those caused by Ptch1 heterozygosity. SMO activation also increased the mammosphere-forming efficiency of primary mammary epithelial cells. However, limiting-dilution transplantation showed a decrease in the frequency of regenerative stem cells in MMTV-SmoM2 epithelium relative to wild type, suggesting enhanced mammosphere-forming efficiency was due to increased survival or activity of division-competent cell types under anchorage-independent growth conditions, rather than an increase in the proportion of regenerative stem cells per se. In human clinical samples, altered hedgehog signaling occurs early in breast cancer development, with PTCH1 expression reduced in approximately 50% of ductal carcinoma in situ (DCIS) and invasive breast cancers (IBC). Conversely, SMO is ectopically expressed in 70% of DCIS and 30% of IBC. Surprisingly, in both human tumors and MMTV-SmoM2 mice, SMO rarely colocalized with the Ki67 proliferation marker. Our data suggest that altered hedgehog signaling may contribute to breast cancer development by stimulating proliferation, and by increasing the pool of division-competent cells capable of anchorage-independent growth.     

Cross-domain retinopathy classification with optical coherence tomography images via a novel deep domain adaptation method

Deep learning based retinopathy classification with optical coherence tomography (OCT) images has recently attracted great attention. However, existing deep learning methods fail to work well when training and testing datasets are different due to the general issue of domain shift between datasets caused by different collection devices, subjects, imaging parameters, etc. To address this practical and challenging issue, we propose a novel deep domain adaptation (DDA) method to train a model on a labeled dataset and adapt it to an unlabelled dataset (collected under different conditions). It consists of two modules for domain alignment, that is, adversarial learning and entropy minimization. We conduct extensive experiments on three public datasets to evaluate the performance of the proposed method. The results indicate that there are large domain shifts between datasets, resulting a poor performance for conventional deep learning methods. The proposed DDA method can significantly outperform existing methods for retinopathy classification with OCT images. It achieves retinopathy classification accuracies of 0.915, 0.959 and 0.990 under three cross-domain (cross-dataset) scenarios. Moreover, it obtains a comparable performance with human experts on a dataset where no labeled data in this dataset have been used to train the proposed DDA method. We have also visualized the learnt features by using the t-distributed stochastic neighbor embedding (t-SNE) technique. The results demonstrate that the proposed method can learn discriminative features for retinopathy classification.     

Differentiating smokers and nonsmokers based on Raman spectroscopy of oral fluid and advanced statistics for forensic applications

Raman spectroscopy has proven to be a valuable tool for analyzing various types of forensic evidence such as traces of body fluids. In this work, Raman spectroscopy was employed as a nondestructive technique for the analysis of dry traces of oral fluid to differentiate between smoker and nonsmoker donors with the aid of advanced statistical tools. A total of 32 oral fluid samples were collected from donors of differing gender, age and race and were subjected to Raman spectroscopic analysis. A genetic algorithm was used to determine eight spectral regions that contribute the most to the differentiation of smokers and nonsmokers. Thereafter, a classification model was developed based on the artificial neural network that showed 100% accuracy after external validation. The developed approach demonstrates great potential for the differentiation of smokers and nonsmokers based on the analysis of dry traces of oral fluid.     

Determination of β-lactam antibiotics in milk based on magnetic molecularly imprinted polymer extraction coupled with liquid chromatography-tandem mass spectrometry

In this work, a rapid and selective method was successfully developed using the magnetic molecularly imprinted polymer (MMIP) as sorbent for the extraction of β-lactam antibiotics (BLAs) from milk samples. The MMIP has been prepared using penicillin V potassium (PENV) as template molecule, methacrylic acid as functional monomer, ethylene glycol dimethacrylate as crosslinking agent and Fe(3)O(4) magnetite as magnetic component. The experimental results showed that the MMIP had high affinity and selectivity toward PENV and other structurally related BLAs. The extraction process was carried out in a single step by mixing the extraction solvent, MMIPs and milk samples under ultrasonic action. When the extraction was completed, the MMIPs adsorbing the analytes were separated from the sample matrix by an external magnet. The analytes eluted from the MMIP were analyzed by liquid chromatography-tandem mass spectrometry. For achieving optimal preconcentration and reducing non-specific interactions, various parameters affecting the extraction efficiency such as extraction mode, extraction solvent, the amount of MMIPs, extraction time, washing solution and eluting solution were comprehensively evaluated. Under the optimal conditions, the detection limits of BLAs are in the range of 1.6-2.8 ng mL(-1). The relative standard deviations of intra- and inter-day ranging from 3.2% to 8.3% and from 3.6% to 9.8% are obtained, respectively. The method was applied to determine BLAs including PENV, amoxicillin and oxacillin in five milk samples from different provenances. The recoveries of BLAs in these samples from 71.6% to 90.7% are obtained.     

Immunopositivity for Siglec-15 in gastric cancer and its association with clinical and pathological parameters

The sialic acid-binding immunoglobulin-type lectin Siglec-15 is a promising target for cancer immunotherapy in several tumor types. The present study aimed to investigate Siglec-15 expression in gastric cancer (GC) patient tissues and to evaluate its clinical value. Siglec-15 expression was evaluated by immunohistochemistry in 71 patients. Siglec-15 staining was observed in tumor cells of 53 (74.64%) patients, with significant association with histologic classification and angiolymphatic invasion (p<0.05). Immunohistochemistry analysis also detected Siglec-15 in tumor-associated stroma cells (macrophages/myeloid cells). There was no significant association with outcome parameters. Siglec-15 expression in well differentiated histological GC tissues and in the tumor microenvironment are potential targets to be further investigated as a novel prognostic factor for GC.Key words: Siglec, sialic acid, immunohistochemistry, clinical-pathological analyses     

偏振光成像技术用于肿瘤病变检测的研究进展

偏振光成像是一种非标记、无损伤检测技术,它与现有非偏振光学方法硬件兼容,但能提供更丰富的样品结构和光学信息,并且对亚波长微观结构变化十分敏感.最近,偏振光成像方法在生物医学,特别是肿瘤癌症检测领域显示出很好的应用前景.本文介绍了常用的偏振光散射成像方法,包括偏振差、偏振度、旋转线偏振成像、偏振显微、穆勒矩阵成像等,并展示这些偏振方法在生物医学领域,特别是癌症检测方面的最新研究进展.目前偏振差、偏振度等成像方法已被初步用于皮肤癌的诊断,而穆勒矩阵包含更为丰富的组织微观结构信息,因而具有更好的诊断应用前景.通过对穆勒矩阵进行分解、变换等处理,可获得具有明确物理意义的成像参数,并发展为针对不同应用的特异性方法.目前,随着新型光源、偏振器件和探测器的出现,特别是数据计算处理能力的急剧提升,偏振成像在数据解释和测量方法方面的研究快速发展,已经在生物医学领域显示出很好的应用前景.