Polarimetric imaging for cervical pre-cancer screening aided by machine learning: ex vivo studies.

Significance: Wide-field imaging Mueller polarimetry is an optical imaging technique that has great potential to become a reliable, fast, non-contact in vivo imaging modality for the early detection of, e.g., deceases and tissue structural malformations, such as cervical intraepithelial neoplasia, in both clinical and low-resource settings. On the other hand, machine learning methods have established themselves as a superior solution in image classification and regression tasks. We combine Mueller polarimetry and machine learning, critically assess the data/classification pipeline, investigate the bias arising from training strategies, and demonstrate how higher levels of detection accuracy can be achieved. Aim: We aim to automate/assist with diagnostic segmentation of polarimetric images of uterine cervix specimens. Approach: A comprehensive capture-to-classification pipeline is developed in house. Specimens are acquired and measured with imaging Mueller polarimeter and undergo histopathological classification. Subsequently, a labeled dataset is created within tagged regions of either healthy or neoplastic cervical tissues. Several machine learning methods are trained utilizing different training-test-set-split strategies, and their corresponding accuracies are compared. Results: Our results include robust measurements of model performance with two approaches: a 90:10 training-test-set-split and leave-one-out cross-validation. By comparing the classifier's accuracy directly with the ground truth obtained during histology analysis, we demonstrate how conventionally used shuffled split leads to an over-estimate of true classifier performance (0.964±0.00). The leave-one-out cross-validation, however, leads to more accurate performance (0.812±0.21) with respect to newly obtained samples that were not used to train the models. Conclusions: Combination of Mueller polarimetry and machine learning is a powerful tool for the task of screening for pre-cancerous conditions in cervical tissue sections. Nevertheless, there is a inherent bias with conventional processes that can be addressed using more conservative classifier training approaches. This results in overall improvements of the sensitivity and specificity of the developed techniques for "unseen" images.

Liquid-crystal based drift-free polarization modulators: Part II. Ultra-stable Stokes and Mueller polarimeters.

We have previously reported a new design for drift-free liquid-crystal polarization modulators (LCMs) based on liquid-crystal variable retarders (LCVRs). Here, we study their performance on Stokes and Mueller polarimeters. LCMs have polarimetric responses similar to LCVRs and can be used as temperature-stable alternatives to many LCVR-based polarimeters. We have built an LCM-based polarization state analyzer (PSA) and compared its performance to an equivalent LCVR-based PSA. Our system parameters remained stable over a wide range of temperature, precisely from 25°C to 50°C. Accurate Stokes and Mueller measurements have been conducted, paving the way to calibration-free polarimeters for demanding applications.

Depolarization imaging for fast and non-invasive monitoring of cervical microstructure remodeling in vivo during pregnancy.

The cervix plays a crucial role in conception, maintenance of pregnancy, and childbirth. The mechanical properties of a pregnant woman's cervix change dramatically during gestation due to a remodeling of its microstructure, necessary for delivery. However, external factors can accelerate this process and lead to prematurity, the primary cause of perinatal mortality worldwide, due to the inefficiency of existing diagnostic methods. This study shows that polarized light is a powerful tool to probe the cervical microstructure during pregnancy. A wide-field multispectral polarimetric imaging system was fabricated to explore in vivo the cervix of full-term pregnant women. The polarimetric properties of the cervix change significantly with pregnancy progression. In particular, a set of several depolarization parameters (intrinsic and extrinsic) showed a strong linear correlation with gestational age in the red part of the visible spectral range. This trend can be attributed, among other things, to a decrease in collagen density and an increase in hydration of cervical connective tissue. Wide field depolarization imaging is a very promising tool for rapid and non-invasive analysis of cervical tissue in vivo to monitor the steady progression of pregnancy, providing the practitioner with useful information to improve the detection of preterm birth.

Liquid-crystal based drift-free polarization modulators: Part I. Design and operation.

We report a new design for temperature-stable polarization modulators. Each modulator is composed of two liquid crystal variable retarders (LCVRs) positioned in such a way that their temperature drifts mutually compensate. We propose a model for the temperature-dependent polarization response of LCVRs, which permits us to establish expressions for the operating point of the system and for its accessible retardance range. We have validated such a model experimentally by thorough analyses of LCVR temperature responses, and we have built a polarization modulator that is stable over a wide range of temperature with commercially available LCVRs.

Monitoring subcutaneous tumors using Mueller polarimetry: study on two types of tumors.

A better understanding of tumor development is crucial for treating cancer. Polarimetric imaging is an interesting alternative for monitoring subcutaneous tumors as it is non-invasive. In this study, a Mueller spectro-polarimeter is used to monitor tumor development on mice injected with non-pigmented breast cancer cells or with pigmented murine melanoma cells. Three stages of non-pigmented tumor development are revealed with three polarimetric parameters. These stages also appear for pigmented tumors, although less clearly. A halo of high depolarization surrounding the non-pigmented tumor in the first stage allows the outlining of the tumor. Considering polarimetric parameters, a biological interpretation is proposed.

Revisiting the generalized polar decomposition of Mueller matrices.

Mueller polarimetry is a powerful imaging modality that has been successfully applied to various application fields. Decomposition of Mueller matrices in elementary components is classically considered in order to unfold complex physical phenomena taking place in probed samples or scenes. In this context, the generalized polar decomposition, also known as Lu and Chipman decomposition, plays a prominent role. In this paper, we show that the set of candidate generalized polar decompositions is richer than the set used so far. Negative-determinant Mueller matrices are naturally addressed in the proposed framework. We show that taking into account those supplementary polar decompositions addresses issues raised in the literature. Application is carried out on synthetic and on measured Mueller matrices.

Innovative integrated numerical-experimental method for high-performance multispectral Mueller polarimeters based on ferroelectric liquid crystals.

In this work, an original and effective integrated numerical-experimental approach is proposed for building a high-performance multispectral Mueller polarimeter based on ferroelectric liquid crystals (FLCs). This method relies on accurate experimental characterization of the optical components specifically selected to construct such a system, combined with a numerical procedure used to optimize it, in the spectral range of interest, by means of a global optimization function. The proposed strategy enabled the construction of an FLC-based Mueller polarimeter in transmission configuration operating between 450 and 700 nm. The robustness of this system to various optical component misalignments, as well as the conditions to keep the measurement error less than 1% over the whole spectral range of interest, have been determined experimentally. The proposed strategy is very well suited to build optimized multispectral Mueller polarimetric systems for biomedical applications for which variations of the order of a few percent in the elements of the measured Mueller matrices need to be appreciated.

Polarimetric measurement utility for pre-cancer detection from uterine cervix specimens.

Prior work demonstrated significant contrast in visible wavelength Mueller matrix images for healthy and pre-cancerous regions of excised cervical tissue. This work demonstrates post-processing compressions of the full Mueller matrix that preserve detection performance. The purpose of this post-processing is to understand polarimetric measurement utility for computing mathematical observers and designing future imaging protocols. The detection performance of the full Mueller matrix, and both linear and non-linear parameters of the Mueller matrix will be compared. The area under the receiver operating characteristic (ROC) curve, otherwise known as the AUC, is the gold standard metric to quantify detection performance in medical applications. An AUC = 1 is perfect detection and AUC = 0.5 is the performance of guessing. Either the scalar retardance or the 3 smallest eigenvalues of the coherency matrix yield an average AUC of 0.94 or 0.93, respectively. When these four non-linear parameters are used simultaneously the average AUC is 0.95. The J-optimal Channelized Quadratic Observer (J-CQO) method for optimizing polarimetric measurements demonstrates equivalent AUC values for the full Muller matrix and 6 J-CQO optimized measurements. The advantage of this optimization is that only 6 measurements, instead of 16 for the full Mueller matrix, are required to achieve this AUC.

Mueller polarimetric imaging of biological tissues: classification in a decision-theoretic framework.

Mueller polarimetry is increasingly recognized as a powerful modality in biomedical imaging. Nevertheless, principled statistical analysis procedures are still lacking in this field. This paper presents a complete pipeline for polarimetric bioimages, with an application to ex vivo cervical precancer detection. In the preprocessing stage, we evaluate the replacement of pixels by superpixels. In the analysis stage, we resort to decision theory to select and tune a classifier. Performances of the retained classifier are evaluated. Decision theory provides a rigorous and versatile framework, allowing generalization to other pathologies, to other imaging procedures, and to classification problems involving more than two classes.

In vivo imaging of uterine cervix with a Mueller polarimetric colposcope.

Mueller polarimetric imaging enables the detection and quantification of modifications of the collagen fibers in the uterine cervix due to the development of a precancerous lesion. This information is not accessible through the use of the classic colposcope, a low magnification microscope used in current practice for cervical cancer screening. However, the in vivo application of Mueller polarimetric imaging poses an instrumental challenge: the device should be sufficiently compact, while still being able to perform fast and accurate acquisition of Mueller matrices in real-world conditions. In this study, the first wide field Mueller Polarimetric Colposcope (MPC) for the in vivo analysis of uterine cervix is presented. The MPC has been fabricated by grafting a miniaturized Mueller polarimetric imager on a classic colposcope. This new imaging tool performs the fast acquisition of Mueller polarimetric images, thus eliminating any blurring effects due to patient movements. It can be easily used by a practitioner with little change to their existing practice. Finally, the MPC was tested in vivo on a number of patients in the field.

Ex vivo Mueller polarimetric imaging of the uterine cervix: a first statistical evaluation.

Early detection through screening plays a major role in reducing the impact of cervical cancer on patients. When detected before the invasive stage, precancerous lesions can be eliminated with very limited surgery. Polarimetric imaging is a potential alternative to the standard screening methods currently used. In a previous proof-of-concept study, significant contrasts have been found in polarimetric images acquired for healthy and precancerous regions of excised cervical tissue. To quantify the ability of the technique to differentiate between healthy and precancerous tissue, polarimetric images of seventeen cervical conization specimens (cone-shaped or cylindrical wedges from the uterine cervix) are compared with results from histopathological diagnoses, which is considered to be the "gold standard." The sensitivity and specificity of the technique are calculated for images acquired at wavelengths of 450, 550, and 600 nm, aiming to differentiate between high-grade cervical intraepithelial neoplasia (CIN 2-3) and healthy squamous epithelium. To do so, a sliding threshold for the scalar retardance parameter was used for the sample zones, as labeled after histological diagnosis. An optimized value of ∼83% is achieved for both sensitivity and specificity for images acquired at 450 nm and for a threshold scalar retardance value of 10.6 deg. This study paves the way for an application of polarimetry in the clinic.

Multimodal nonlinear optical microscopy with shaped 10 fs pulses.

We demonstrate the use of shaped 10 fs pulses for multimodal microscopy. The combination of a broadband oscillator and a pulse shaper provides a flexible light source that can be optimized for various nonlinear effects produced in the sample, either for signal intensity or for selectivity. While the highest nonlinear generation efficiency is achieved with the shortest pulses, more complex waveforms address specific transitions in the sample for better contrast. This is shown experimentally with the imaging of a moss leaf and of human skin biopsies using coherent anti-Stokes Raman scattering, two-photon fluorescence and second harmonic generation signals.

Mapping impurity of single-walled carbon nanotubes in bulk samples with multiplex coherent anti-stokes Raman microscopy.

Mapping of defects in bulk samples of single-walled carbon nanotubes (SWNT) is performed via multiplex coherent anti-Stokes Raman microscopy. The D and G vibrational bands are acquired simultaneously, and their relative amplitude is used as a criterion to quantify the local purity in spin-coated SWNT samples. We observe that defects induced by oxidation are related to the spatial dispersion of nanotubes in a solid distribution.

Full characterization of the third-order nonlinear susceptibility using a single-beam coherent anti-Stokes Raman scattering setup.

A method for the full determination of the third-order nonlinear coherent anti-Stokes Raman scattering (CARS) susceptibility is presented, which relies on phase control of a single ultrabroadband femtosecond laser pulse. A narrowband phase gate is scanned to perform double quadrature spectral interferometry, which reveals amplitude and phase of a multiplex CARS field. A single calibration measurement in a nonresonant sample allows for the characterization of the susceptibility in amplitude and phase. This scheme is demonstrated experimentally for the fingerprint region of toluene.

Multiplex coherent anti-Stokes Raman microspectroscopy with tailored Stokes spectrum.

We combined the ultrabroadband supercontinuum of a photonic crystal fiber with a pulse shaper, resulting in a highly flexible light source for multiplex coherent anti-Stokes Raman microscopy. Implemented as the Stokes pulse, it provides tailored selection of the relevant Raman transitions, resulting in a reduced photon load and partial suppression of the nonresonant background. This experiment exploits the advantages of multiplex excitation with the increased acquisition speed of single-channel detection. The molecule-specific Stokes pulses are demonstrated for chemical mapping of a polymer blend.