Deep learning-based classification of dermatological lesions given a limited amount of labeled data

BACKGROUND

Artificial intelligence (AI) techniques are promising in early diagnosis of skin diseases. However, a precondition for their success is the access to large-scaled annotated data. Until now, obtaining this data has only been feasible with very high personnel and financial resources.

OBJECTIVES

The aim of this study was to overcome the obstacle caused by the scarcity of labeled data.

METHODS

To simulate the scenario of label shortage, we discarded a proportion of labels of the training set. The training set consisted of both labeled and unlabeled images. We then leveraged a self-supervised learning technique to pre-train the AI model on the unlabeled images. Next, we fine-tuned the pre-trained model on the labeled images.

RESULTS

When the images in the training dataset were fully labeled, the self-supervised pre-trained model achieved 95.7% of accuracy, 91.7% of precision and 90.7% of sensitivity. When only 10% of the data was labeled, the model could still yield 87.7% of accuracy, 81.7% of precision and 68.6% of sensitivity. In addition, we also empirically verified that the AI model and dermatologists are consistent in visually inspecting the skin images.

CONCLUSIONS

The experimental results demonstrate the great potential of the self-supervised learning in alleviating the scarcity of annotated data.

Spectral cross-cumulants for multicolor super-resolved SOFI imaging

Super-resolution optical fluctuation imaging provides a resolution beyond the diffraction limit by analysing stochastic fluorescence fluctuations with higher-order statistics. Using nth order spatio-temporal cross-cumulants the spatial resolution and the sampling can be increased up to n-fold in all spatial dimensions. In this study, we extend the cumulant analysis into the spectral domain and propose a multicolor super-resolution scheme. The simultaneous acquisition of two spectral channels followed by spectral cross-cumulant analysis and unmixing increases the spectral sampling. The number of discriminable fluorophore species is thus not limited to the number of physical detection channels. Using two color channels, we demonstrate spectral unmixing of three fluorophore species in simulations and experiments in fixed and live cells. Based on an eigenvalue/vector analysis, we propose a scheme for an optimized spectral filter choice. Overall, our methodology provides a route for easy-to-implement multicolor sub-diffraction imaging using standard microscopes while conserving the spatial super-resolution property.

Erratum: Reduction of Abeta amyloid pathology in APPPS1 transgenic mice in the absence of gut microbiota

This corrects the article DOI: 10.1038/srep41802.

Reduction of Abeta amyloid pathology in APPPS1 transgenic mice in the absence of gut microbiota

Alzheimer’s disease is the most common form of dementia in the western world, however there is no cure available for this devastating neurodegenerative disorder. Despite clinical and experimental evidence implicating the intestinal microbiota in a number of brain disorders, its impact on Alzheimer’s disease is not known. To this end we sequenced bacterial 16S rRNA from fecal samples of Aβ precursor protein (APP) transgenic mouse model and found a remarkable shift in the gut microbiota as compared to non-transgenic wild-type mice. Subsequently we generated germ-free APP transgenic mice and found a drastic reduction of cerebral Aβ amyloid pathology when compared to control mice with intestinal microbiota. Importantly, colonization of germ-free APP transgenic mice with microbiota from conventionally-raised APP transgenic mice increased cerebral Aβ pathology, while colonization with microbiota from wild-type mice was less effective in increasing cerebral Aβ levels. Our results indicate a microbial involvement in the development of Abeta amyloid pathology, and suggest that microbiota may contribute to the development of neurodegenerative diseases.

Anisotropic X-Ray Dark-Field Tomography: A Continuous Model and its Discretization

The x-ray dark-field signal measured in grating interferometers is anisotropic, depending on both the beam direction and the grating orientation with respect to the sample. We present a novel general closed-form, continuous forward model of the anisotropic dark-field signal. Furthermore, we derive a discretization using spherical harmonics, leading to a large-scale linear inverse problem. We present first experimental results of a wooden sample, demonstrating marked advantages over previous results, in particular, the resolution of multiple scattering directions in one volume element.

In vivo imaging of murine endocrine islets of Langerhans with extended-focus optical coherence microscopy

AIMS/HYPOTHESIS

Structural and functional imaging of the islets of Langerhans and the insulin-secreting beta cells represents a significant challenge and a long-lasting objective in diabetes research. In vivo microscopy offers a valuable insight into beta cell function but has severe limitations regarding sample labelling, imaging speed and depth, and was primarily performed on isolated islets lacking native innervations and vascularisation. This article introduces extended-focus optical coherence microscopy (xfOCM) to image murine pancreatic islets in their natural environment in situ, i.e. in vivo and in a label-free condition.

METHODS

Ex vivo measurements on excised pancreases were performed and validated by standard immunohistochemistry to investigate the structures that can be observed with xfOCM. The influence of streptozotocin on the signature of the islets was investigated in a second step. Finally, xfOCM was applied to make measurements of the murine pancreas in situ and in vivo.

RESULTS

xfOCM circumvents the fundamental physical limit that trades lateral resolution for depth of field, and achieves fast volumetric imaging with high resolution in all three dimensions. It allows label-free visualisation of pancreatic lobules, ducts, blood vessels and individual islets of Langerhans ex vivo and in vivo, and detects streptozotocin-induced islet destruction.

CONCLUSIONS/INTERPRETATION

Our results demonstrate the potential value of xfOCM in high-resolution in vivo studies to assess islet structure and function in animal models of diabetes, aiming towards its use in longitudinal studies of diabetes progression and islet transplants.

Laser Doppler imaging for intraoperative human brain mapping

The identification and accurate location of centers of brain activity are vital both in neuro-surgery and brain research. This study aimed to provide a non-invasive, non-contact, accurate, rapid and user-friendly means of producing functional images intraoperatively. To this end a full field Laser Doppler imager was developed and integrated within the surgical microscope and perfusion images of the cortical surface were acquired during awake surgery whilst the patient performed a predetermined task. The regions of brain activity showed a clear signal (10-20% with respect to the baseline) related to the stimulation protocol which lead to intraoperative functional brain maps of strong statistical significance and which correlate well with the preoperative fMRI and intraoperative cortical electro-stimulation. These initial results achieved with a prototype device and wavelet based regressor analysis (the hemodynamic response function being derived from MRI applications) demonstrate the feasibility of LDI as an appropriate technique for intraoperative functional brain imaging.

Image formation in fluorescence coherence-gated imaging through scattering media

Recently, we have experimentally demonstrated a new form of cross-sectional, coherence-gated fluorescence imaging referred to as SD-FCT ('spectral-domain fluorescence coherence tomography'). Imaging in SD-FCT is accomplished by spectrally detecting self-interference of the spontaneous emission of fluorophores, thereby providing depth-resolved information on the axial positions of fluorescent probes. Here, we present a theoretical investigation of the factors affecting the detected SD-FCT signal through scattering media. An imaging equation for SD-FCT is derived that includes the effects of defocusing, numerical-aperture, and the optical properties of the medium. A comparison between the optical sectioning capabilities of SD-FCT and confocal microscopy is also presented. Our results suggest that coherence gating in fluorescence imaging may provide an improved approach for depth-resolved imaging of fluorescently labeled samples; high axial resolution (a few microns) can be achieved with low numerical apertures (NA<0.09) while maintaining a large depth of field (a few hundreds of microns) in a relatively low scattering medium (6 mean free paths), whereas moderate NA's can be used to enhance depth selectivity in more highly scattering biological samples.

Extended focus depth for Fourier domain optical coherence microscopy

We report on a new detection scheme for Fourier domain optical coherence microscopy that exhibits high transverse resolution along an axially extended focal range. Nearly constant transverse resolution of approximately 1.5 microm along a focal range of 200 microm is experimentally verified with a maximum sensitivity of 105 dB. A broad-bandwidth Ti:sapphire laser allowed for an axial resolution of 3 microm in air.

High light field confinement for fluorescent correlation spectroscopy using a solid immersion lens

In this paper we present recent single molecule detection experiment using a solid immersion lens (SIL) for fluorescent correlation spectroscopy measurements. We compared the performance of the SIL in combination with an air objective (40x, numerical aperture (NA)=1.15) with a water immersion objective (40x, NA=0.6) in a confocal microscope system (ConfoCorr 1). Important parameters for single molecule experiments such as collection efficiency and excitation field confinement were investigated. Although the two set-ups have similar numerical aperture the measurements demonstrated higher field confinement and better collection efficiency for the SIL system in comparison to the conventional confocal set-up. Adding spherical aberrations shifts the sample volume up to 4 microm away from the plane surface of the SIL and conserves a diffraction limited focal volume. In this case the FCS autocorrelation demonstrates a free 3D diffusion of dye molecules in a highly confined light field.

Spatially incoherent illumination as a mechanism for cross-talk suppression in wide-field optical coherence tomography

Comparison of two illumination modes for wide-field optical coherence tomography has revealed that spatially coherent illumination generates coherent cross talk, causing significant image degradation, and that spatially incoherent illumination, with an adequate interferometer design, provides an efficient mechanism for suppression of coherent cross talk. This is shown by comparison of a pulsed laser with a thermal light source for a U.S. Air Force resolution target covered with a scattering solution made from microbeads as well as for an ex vivo tooth.

Self-referenced method for optical path difference calibration in low-coherence interferometry

A simple method for the calibration of optical path difference modulation in low-coherence interferometry is presented. Spectrally filtering a part of the detected interference signal results in a high-coherence signal that encodes the scan imperfections and permits their correction. The method is self-referenced in the sense that no secondary high-coherence light source is necessary. Using a spectrometer setup for spectral filtering allows for flexibility in both the choice of calibration wavelength and the maximum scan range. To demonstrate the method's usefulness, it is combined with a recently published digital spectral shaping technique to measure the thickness of a pellicle beam splitter with a white-light source.

High-speed femtosecond pump-probe spectroscopy with a smart pixel detector array

A new femtosecond pump-probe spectroscopy technique is demonstrated that permits the high-speed, parallel acquisition of pump-probe measurements at multiple wavelengths. This is made possible by use of a novel, two-dimensional smart pixel detector array that performs amplitude demodulation in real time on each pixel. This detector array can not only achieve sensitivities comparable with lock-in amplification but also simultaneously performs demodulation of probe transmission signals at multiple wavelengths, thus permitting rapid time- and wavelength-resolved femtosecond pump-probe spectroscopy. Measurements on a thin sample of bulk GaAs are performed across 58 simultaneous wavelengths. Differential probe transmission changes as small as approximately 2 x 10(-4) can be measured over a 5-ps delay scan in only approximately 3 min. This technology can be applied to a wide range of pump-probe measurements in condensed matter, chemistry, and biology.

Numerical dispersion compensation for Partial Coherence Interferometry and Optical Coherence Tomography

Dispersive samples introduce a wavelength dependent phase distortion to the probe beam. This leads to a noticeable loss of depth resolution in high resolution OCT using broadband light sources. The standard technique to avoid this consequence is to balance the dispersion of the sample byarrangingadispersive materialinthereference arm. However, the impact of dispersion is depth dependent. A corresponding depth dependent dispersion balancing technique is diffcult to implement. Here we present a numerical dispersion compensation technique for Partial Coherence Interferometry (PCI) and Optical Coherence Tomography (OCT) based on numerical correlation of the depth scan signal with a depth variant kernel. It can be used a posteriori and provides depth dependent dispersion compensation. Examples of dispersion compensated depth scan signals obtained from microscope cover glasses are presented.

Actively mode-locked visible upconversion fiber laser

Active mode locking of a Pr(3+)/Yb (3+) -doped upconversion fluoride fiber laser with an all-fiber ZnO acousto-optic phase modulator is demonstrated for the first time to the authors' knowledge. Optical pulses of ~550-ps duration with a repetition rate of 239 MHz at a wavelength of 635 nm have been generated.

Corneal Morphology in vitro After Superficial Keratectomy With Q-switched Er:YSGG and Free-running Er:YAG Lasers

PURPOSE

Examination of morphology in corneal ablation induced by a q-switched Er:YSGG (2.79 microm) laser and a free-running Er:YAG laser (2.94 microm).

METHODS

Defined ablation of 6-mm diameter and 15, 30, 55, 90, and 120-microm depth was performed on freshly enucleated swine eyes. Er:YSGG laser parameters: fluence 1.6 J/cm2, frequency 6 Hz, spot-size 465 microm FWHM, scanning-mode, pulse number 1520 to 6210. Er:YAG laser parameters: fluence 3 J/cm2, frequency 1.5 Hz, spotsize 6 mm, wide area ablation, pulse number 2 to 13. Corneal morphology was analyzed by gross photography, histology, scanning electron microscopy, and scanning nearfield acoustic microscopy.

RESULTS

Histology showed thermal damage of 5 to 15 microm in depth caused by the Er:YSGG laser in comparison with 10 to 20 microm by the Er:YAG laser. Average roughness of the ablated surface measured with scanning nearfield acoustic microscopy was 20 to 40 microm for the Er:YSGG laser and 5 to 15 microm for the Er:YAG laser. These data confirm the subjective impression of images created by scanning electron microscopy and gross photography.

CONCLUSIONS

Although the Er:YAG laser system appeared to demonstrate a smoother corneal surface than the q-switched Er:YSGG laser, the thermal damage in either case poses a potential limitation for clinical use in lamellar refractive surgery.

[Decrease of retinal image contrast after photorefractive keratectomy, improvement within the scope of surface restitution]

Photorefractive keratectomy (PRK) with the ArF excimer lasers in current use usually approximates the intended corneal curvature by a mean of a delicate step-type pattern that is smooth off afterwards by reepithelialization and tear film. The present study was based on a model eye with axial myopia of -6 D but otherwise the optical and geometric properties of the Gullstrand model eye and was designed to investigate to what extent. (1) corneal step patterns can reduce retinal image contrast and (2) smoothing effects can restore such a loss. METHODS. The corneal surface resulting from PRK in the case of a myopia of -6 D (optical zone diameter 6 mm) is calculated for the parameters of the model eye. The retinal image contrasts of bar patterns are calculated by PSF (point spread function) analysis: varying size of pupil, wavelength, bar width, ablation step height and degree of smoothing. RESULTS. Step height influences retinal image contrast crucially. With step heights above 0.4 micron a massive loss of retinal image contrast must be expected, which can, however, be corrected to a useful extent by surface-smoothing effects. CONCLUSION. This study indicates that PRK with excimer lasers should be performed with low fluence and correspondingly low corneal step heights.

Mathematical Simulation of Retinal Image Contrast After Photorefractive Keratectomy With a Diaphragm Mask

BACKGROUND

Photorefractive keratectomy (PRK) using a dilating diaphragm mask engraves a delicate three-dimensional staircase pattern into a formerly smooth corneal surface. The created steps are later smoothed by tear film and wound healing processes. The present study investigates, in a mathematical simulation, the effects that such staircase patterns and their smoothing may have on retinal image contrast.

METHODS

All simulations are based on the Gullstrand eye model and calculate retinal image contrast from point spread function (PSF) analysis of Gullstrand eyes treated by simulated PRK under various conditions.

RESULTS

The simulations indicate that PRK can reduce retinal contrast markedly. The most critical factor for such a reduction is the step height of the ablation pattern. With step heights below 0.4 microns, loss of contrast due to the created staircase pattern is always moderate and should be restored during early wound healing. Complete wound healing may smooth out larger step heights. Micromovements during PRK also can lead to partial loss of retinal image contrast.

CONCLUSIONS

Simulation of retinal contrast after PRK shows that step heights below 0.4 microns seem to be acceptable. A minimization of the micromovements during PRK can offset some of the reduction of retinal contrast.

Kinetics and dynamics of sematilide

Sematilide HCl is a novel class III antiarrhythmic drug. The goals of this study in volunteers were to determine the pharmacokinetics, effect (QTc interval), and tolerability after intravenous and oral administration of 25 mg of the drug. Plasma and urine concentrations were measured by a specific high-performance liquid chromatography method. Pharmacokinetic data analysis used a compartment model independent approach. An effect on QTc was observed only after intravenous administration, and its relationship to the plasma concentration showed a counterclockwise hysteresis. A semiparametric approach was used to collapse the hysteresis and then evaluate the effect-site-concentration-to-effect relationship. After intravenous and oral administration, 75.1 (6.5)% (mean +/- SD) and 36.0 (11.5)% of the dose was excreted unchanged in urine, respectively. The respective renal clearances were 250 (41) ml.min-1 and 222 (44) ml.min-1. The bioavailability of sematilide was 0.47 (0.15). A maximum percent effect on QTc of 12 (1)% occurred with a delay of 14 min after termination of an intravenous infusion of 10 min. After collapsing the hysteresis, the pharmacokinetic-pharmacodynamic data could be fitted appropriately by a linear model in four subjects and by an Emax model in two subjects. Sematilide HCl was well tolerated.

[Photocoagulation in the edematous and non-edematous retina with the cw-laser of different wavelengths]

The impact of wavelength upon the laser power required in photocoagulation of oedematous and non-oedematous retinal areas was investigated in a restricted clinical study. The following laser systems were applied: an argon laser (514 nm), a diode-pumped frequency-doubled Nd:YAG laser (532 nm) and a diode laser (810 nm). All lasers had broadly similar properties as regards beam characteristics, beam conduction and application optics. Twenty-five eyes of 25 patients with diabetic retinopathy were included in the study. Each eye was treated with each laser in areas of edema and no edema. Apart from the wavelength, laser power was the only variable permitted. The power required for photocoagulation at 514 nm and 532 nm was identical under every aspect of the study. In comparison with 514 nm or 532 nm photocoagulations, 810 nm photocoagulations required 6.0 +/- 0.9-fold higher power settings in nonedematous retina and 5.75 +/- 1.0 fold higher power settings in edematous retina. At every wavelength investigated, retinal edema raised the average laser power needed by more than 20%. The extent of this effect seems to decline with longer wavelengths.

Gas thermometry by Fourier analysis of rotational coherent anti-Stokes Raman scattering

Using Fourier analysis, the periodicity of rotational coherent anti-Stokes Raman scattering spectra is shown to offer a new possibility for gas-temperature evaluation. In a simple way, this approach can be used for data reduction leading to a rapid temperature-evaluation procedure.