Marine Amouroux

Classification of ultraviolet irradiated mouse skin histological stages by bimodal spectroscopy: multiple excitation autofluorescence and diffuse reflectance

Histopathological analysis and in vivo optical spectroscopy were used to discriminate several histological stages of UV-irradiated mouse skin. At different times throughout the 30-week irradiation, autofluorescence (AF) and diffuse reflectance (DR) spectra were acquired in a bimodal approach. Then skin was sampled and processed to be classified, according to morphological criteria, into four histological categories: normal, and three types of hyperplasia (compensatory, atypical, and dysplastic). After extracting spectral characteristics, principal component analysis (data reduction) and the k-nearest neighbor classifying method were applied to compare diagnostic performances of monoexcitation AF (based on each of the seven excitation wavelengths: 360, 368, 390, 400, 410, 420, and 430 nm), multiexcitation AF (combining the seven excitation wavelengths), DR, and bimodal spectroscopies. Visible wavelengths are the most sensitive ones to discriminate compensatory from precancerous (atypical and dysplastic) states. Multiexcitation AF provides an average 6-percentage-point increased sensitivity compared to the best scores obtained with monoexcitation AF for all pairs of tissue categories. Bimodality results in a 4-percentage-point increase of specificity when discriminating the three types of hyperplasia. Thus, bimodal spectroscopy appears to be a promising tool to discriminate benign from precancerous stages; clinical investigations should be carried out to confirm these results.

Hybrid feature selection and SVM-based classification for mouse skin precancerous stages diagnosis from bimodal spectroscopy

This paper deals with multi-class classification of skin pre-cancerous stages based on bimodal spectroscopic features combining spatially resolved AutoFluorescence (AF) and Diffuse Reflectance (DR) measurements. A new hybrid method to extract and select features is presented. It is based on Discrete Cosine Transform (DCT) applied to AF spectra and on Mutual Information (MI) applied to DR spectra. The classification is performed by means of a multi-class SVM: the M-SVM2. Its performance is compared with the one of the One-Versus-All (OVA) decomposition method involving bi-class SVMs as base classifiers. The results of this study show that bimodality and the choice of an adequate spatial resolution allow for a significant increase in diagnostic accuracy. This accuracy can get as high as 81.7% when combining different distances in the case of bimodality.

A safe, low-cost, and portable instrumentation for bedside time-resolved picosecond near infrared spectroscopy

Continuous wave Near InfraRed Spectroscopy (NIRS) has been used successfully in clinical environments for several years to detect cerebral activation thanks to oxymetry (i.e. absorption of photons by oxy- and deoxy- hemoglobin) measurement. The goal of our group is to build a clinically-adapted time-resolved NIRS setup i.e. a setup that is compact and robust enough to allow bedside measurements and that matches safety requirements with human patients applications. Indeed our group has already shown that time resolution allows spatial resolution and improves sensitivity of cerebral activation detection. The setup is built with four laser diodes (excitation wavelengths: 685, 780, 830 and 870 nm) whose emitted light is injected into four optical fibers; detection of reflected photons is made through an avalanche photodiode and a high resolution timing module used to record Temporal Point Spread Functions (TPSF). Validation of the device was made using cylindrically-chaped phantoms with absorbing and/or scattering inclusions. Results show that recorded TPSF are typical both of scattering and absorbing materials thus demonstrating that our apparatus would detect variation of optical properties (absorption and scattering) deep within a diffusive media just like a cerebral activation represents a rise of absorption in the cortex underneath head surface.

Robust bladder image registration by redefining data-term in total variational approach

Cystoscopy is the standard procedure for clinical diagnosis of bladder cancer diagnosis. Bladder carcinoma in situ are often multifocal and spread over large areas. In vivo, localization and follow-up of these tumors and their nearby sites is necessary. But, due to the small field of view (FOV) of the cystoscopic video images, urologists cannot easily interpret the scene. Bladder mosaicing using image registration facilitates this interpretation through the visualization of entire lesions with respect to anatomical landmarks. The reference white light (WL) modality is affected by a strong variability in terms of texture, illumination conditions and motion blur. Moreover, in the complementary fluorescence light (FL) modality, the texture is visually different from that of the WL. Existing algorithms were developed for a particular modality and scene conditions. This paper proposes a more general on fly image registration approach for dealing with these variability issues in cystoscopy. To do so, we present a novel, robust and accurate image registration scheme by redefining the data-term of the classical total variational (TV) approach. Quantitative results on realistic bladder phantom images are used for verifying accuracy and robustness of the proposed model. This method is also qualitatively assessed with patient data mosaicing for both WL and FL modalities.

Towards skin image mosaicing

This paper presents a framework for mosaicing high resolution skin video sequences in the context of teleder-matology. While considering different stages of the mosaicing pipeline, including stitching and blending, several feature- and intensity-based image registration approaches are compared. Their performances in terms of quantitative and qualitative results are discussed so as to move towards the selection of the most suited approach. Although the intensity based approach proved to be more precise over short displacements, the feature based approach is advantageous in terms of computation time apart from being more reliable over large displacements, thus permitting a faster mosaic construction by skipping over some frames in the sequence.

Spatially resolved bimodal spectroscopy for classification/evaluation of mouse skin inflammatory and pre-cancerous stages

Spatially-resolved bimodal spectroscopy (multiple AutoFluorescence AF excitation and Diffuse Reflectance DR), was used in vivo to discriminate various healthy and precancerous skin stages in a pre-clinical model (UV-irradiated mouse): Compensatory Hyperplasia CH, Atypical Hyperplasia AH and Dysplasia D. A specific data preprocessing scheme was applied to intensity spectra (filtering, spectral correction and intensity normalization), and several sets of spectral characteristics were automatically extracted and selected based on their discrimination power, statistically tested for every pair-wise comparison of histological classes. Data reduction with Principal Components Analysis (PCA) was performed and 3 classification methods were implemented (k-NN, LDA and SVM), in order to compare diagnostic performance of each method. Diagnostic performance was studied and assessed in terms of Sensibility (Se) and Specificity (Sp) as a function of the selected features, of the combinations of 3 different inter-fibres distances and of the numbers of principal components, such that: Se and Sp ≈ 100% when discriminating CH vs. others; Sp ≈ 100% and Se > 95% when discriminating Healthy vs. AH or D; Sp ≈ 74% and Se ≈ 63% for AH vs. D.

Multimodal fluorescence imaging navigation for surgical guidance of malignant tumors in photosensitized tissues of neural system and other organs

A 5-ALA-induced fluorescence-based imaging device for guidance during surgery of malignant and non-malignant preliminary photosensitized tumors is presented. The setup fits existing clinical optical rigid and flexible endoscopes and operation microscopes. It consists of three light sources including white light, red light fluorescence excitation and blue light fluorescence excitation sources. The light from any combination of the latter sources is delivered to tissue using specially designed fiber optic light guide. Two cameras are used to acquire fluorescence and back reflected white light images: a gray-level camera for fluorescence in the far red range and a color camera for white light images. A dichroic mirror is implemented to spectrally split the light coming from tissue. Images from both cameras are processed into a computer with specially developed software where it can be displayed in different modes including overlaying or been used for image mosaicing which allows for increasing the intrinsic reduced field of view of endoscopes by providing highly resolved extended cartography. Experiments were carried out on phantoms and on patients in clinical conditions during surgery of brain and other tissues. Blue light excitation was more sensitive for thin tumors but red light excitation was more beneficial for solid tumors and for navigation in presence of slight bleeding.

Spatially resolved spectroscopy for guiding margin delineation during human skin carcinomas resection: first clinical results on diffuse reflectance and autofluorescence spectra and in vivo skin optical properties

Optical spectra acquired on skin depend greatly on skin content in chromophores especially on photons absorbers (melanin) and on fluorescent molecules such as collagen and elastin. Such skin content in chromophores varies from one person to another one even within the same phototype class. Therefore, optical spectra must be standardized in order to be as independent as possible of inter-individual variability. Such a standardization should help increase the diagnosis accuracy of optical spectroscopy. In this study, we aim at defining the anatomical site that would allow best standardization. Standardization is evaluated through the ability of spectroscopy to discriminate pairs of histological classes such as BCC, SCC, AK and normal skin. Bimodal spectroscopy is used combining AutoFluorescence (AF) and Diffuse Reflectance (DR). Three anatomical sites are compared for reference spectra acquisition: non-lesional (NL) skin sites (i.e. the same anatomical sites as for BCC, SCC or AK), inner hand and inner wrist. AF raw data show a high (40 %) inter-individual variability of reference spectra intensity acquired on any of the three anatomical sites. Standardization using reference spectra allow best discrimination of histological classes when reference spectra acquired on NL skin sites and on inner hand are used for standardization of AF spectroscopy and DR spectroscopy, respectively.

Image quality assessment for teledermatology: from consumer devices to a dedicated medical device

Aging population as well as growing incidence of type 2 diabetes induce a growing incidence of chronic skin disorders. In the meantime, chronic shortage of dermatologists leaves some areas underserved. Remote triage and assistance to homecare nurses (known as “teledermatology”) appear to be promising solutions to provide dermatological valuation in a decent time to patients wherever they live. Nowadays, teledermatology is often based on consumer devices (digital tablets, smartphones, webcams) whose photobiological and electrical safety levels do not match with medical devices’ levels. The American Telemedicine Association (ATA) has published recommendations on quality standards for teledermatology. This “quick guide” does not address the issue of image quality which is critical in domestic environments where lighting is rarely reproducible. Standardized approaches of image quality would allow clinical trial comparison, calibration, manufacturing quality control and quality insurance during clinical use. Therefore, we defined several critical metrics using calibration charts (color and resolution charts) in order to assess image quality such as resolution, lighting uniformity, color repeatability and discrimination of key couples of colors. Using such metrics, we compared quality of images produced by several medical devices (handheld and video-dermoscopes) as well as by consumer devices (digital tablet and cameras) widely spread among dermatologists practice. Since diagnosis accuracy may be impaired by “low quality-images”, this study highlights that, from an optical point of view, teledermatology should only be performed using medical devices. Furthermore, a dedicated medical device should probably be developed for the time follow-up of skin lesions often managed in teledermatology such as chronic wounds that require i) noncontact imaging of ii) large areas of skin surfaces, both criteria that cannot be matched using dermoscopes.

Multidimensional spectroscopic data fusion improves precancerous tissue discrimination based on spatially resolved autofluorescence and diffuse reflectance spectroscopy

The current study deals with new perspectives to perform more efficient classification of mouse skin precancerous stages by means of a decision fusion scheme based on belief functions and exploiting the spatial resolution of the autofluorescence and diffuse reflectance spectroscopic data.