A comparison between a white LED confocal imaging system and a conventional flash fundus camera using chromaticity analysis.

BACKGROUND: Conventional flash fundus cameras capture color images that are oversaturated in the red channel and washed out in the green and blue channels, resulting in a retinal picture that often looks flat and reddish. A white LED confocal device was recently introduced to provide a high-quality retinal image with enhanced color fidelity. In this study, we aimed to evaluate the color rendering properties of the white LED confocal system and compare them to those of a conventional flash fundus camera through chromaticity analysis. METHODS: A white LED confocal device (Eidon, Centervue, Padova, Italy) and a traditional flash fundus camera (TRC-NW8, Topcon Corporation, Tokyo, Japan) were used to capture fundus images. Color images were evaluated with respect to chromaticity. Analysis was performed according to the image color signature. The color signature of an image was defined as the distribution of its pixels in the rgb chromaticity space. The descriptors used for the analysis are the average and variability of the barycenter positions, the average of the variability and the number of unique colors (NUC) of all signatures. RESULTS: Two hundred thirty-three color photographs were acquired with each retinal camera. The images acquired by the confocal white LED device demonstrated an average barycenter position (rgb = [0.448, 0.328, 0.224]) closer to the center of the chromaticity space, while the conventional fundus camera provides images with a clear shift toward red at the expense of the blue and green channels (rgb = [0.574, 0.278, 0.148] (p < 0.001). The variability of the barycenter positions was higher in the white LED confocal system than in the conventional fundus camera. The average variability of the distributions was higher (0.003 ± 0.007, p < 0.001) in the Eidon images compared to the Topcon camera, indicating a greater richness of color. The NUC percentage was higher for the white LED confocal device than for the conventional flash fundus camera (0.071% versus 0.025%, p < 0.001). CONCLUSIONS: Eidon provides more-balanced color images, with a wider richness of color content, compared to a conventional flash fundus camera. The overall higher chromaticity of Eidon may provide benefits in terms of discriminative power and diagnostic accuracy.

Spectrally resolved autofluorescence imaging in posterior uveitis.

Clinical discrimination of posterior uveitis entities remains a challenge. This exploratory, cross-sectional study investigated the green (GEFC) and red emission fluorescent components (REFC) of retinal and choroidal lesions in posterior uveitis to facilitate discrimination of the different entities. Eyes were imaged by color fundus photography, spectrally resolved fundus autofluorescence (Color-FAF) and optical coherence tomography. Retinal/choroidal lesions' intensities of GEFC (500-560 nm) and REFC (560-700 nm) were determined, and intensity-normalized Color-FAF images were compared for birdshot chorioretinopathy, ocular sarcoidosis, acute posterior multifocal placoid pigment epitheliopathy (APMPPE), and punctate inner choroidopathy (PIC). Multivariable regression analyses were performed to reveal possible confounders. 76 eyes of 45 patients were included with a total of 845 lesions. Mean GEFC/REFC ratios were 0.82 ± 0.10, 0.92 ± 0.11, 0.86 ± 0.10, and 1.09 ± 0.19 for birdshot chorioretinopathy, sarcoidosis, APMPPE, and PIC lesions, respectively, and were significantly different in repeated measures ANOVA (p < 0.0001). Non-pigmented retinal/choroidal lesions, macular neovascularizations, and fundus areas of choroidal thinning featured predominantly GEFC, and pigmented retinal lesions predominantly REFC. Color-FAF imaging revealed involvement of both, short- and long-wavelength emission fluorophores in posterior uveitis. The GEFC/REFC ratio of retinal and choroidal lesions was significantly different between distinct subgroups. Hence, this novel imaging biomarker could aid diagnosis and differentiation of posterior uveitis entities.

Age-Related Changes in Murine Corneal Nerves.

PURPOSE: The aim of this study is to determine age-related morphological changes in the corneal subbasal nerve plexus (SNP) in two inbred mouse strains. MATERIALS AND METHODS: The corneal SNP was investigated by in vivo confocal laser scanning microscopy (CLSM) in 0.5-, 1-, 1.5-, and 2-year-old C57BL/6J mice and in 0.5- and 1-year-old BALB/c mice (n = 4 per age category and strain; 10 images per mouse). Fixed corneal samples from C57BL/6J mice were also analyzed after PGP9.5 staining. Nerve fiber density (NFD) was determined using the semi-automated NeuronJ program. In addition, a new custom-designed, fully automated computerized technique based on oriented multiscale matched filtering was tested to objectify and accelerate image analysis. RESULTS: C57BL/6J mice showed low NFD (11.7 ± 0.5 mm/mm2). Aging from 0.5 to 1, 1.5, and 2 years resulted in significant reductions in subbasal NFD by 34%, 49%, and 66%, respectively. The decline in nerve fibers revealed by in vivo CLSM together with NeuronJ quantification was confirmed by ex vivo immunohistochemical analyses. Subbasal NFD in BALB/c mice (30.0 ± 1.4 mm/mm2) was 3-fold higher than in C57BL/6J mice. Aging from 0.5 to 1 year resulted in a significant 17% reduction in NFD. With the automated approach, NFD of 22.6 ± 2.9 mm/mm2 and a 45% reduction during aging was determined from the same images. CONCLUSIONS: An age-related reduction in subbasal corneal nerve fibers was observed. The differing extent of reduction in the two mouse strains may be accounted for by genetic factors. Automated NFD quantification of corneal nerve fibers in mice appears to be a useful, reliable, objective, and time-saving tool.

Corneal Nerve Regeneration After Collagen Cross-Linking Treatment of Keratoconus: A 5-Year Longitudinal Study.

IMPORTANCE: It is unknown whether a neurotrophic deficit or pathologic nerve morphology persists in keratoconus in the long term after corneal collagen cross-linking (CXL) treatment. Nerve pathology could impact long-term corneal status in patients with keratoconus. OBJECTIVE: To determine whether CXL treatment of keratoconus results in normalization of subbasal nerve density and architecture up to 5 years after treatment. DESIGN, SETTING, AND PARTICIPANTS: Observational study of 19 patients with early-stage keratoconus indicated for a first CXL treatment with longitudinal follow-up to 5 years postoperatively (examinations were performed from 2009 to 2015; analysis was performed from February to May 2015) and 19 age-matched healthy volunteers at a primary care center and a university hospital ophthalmology department. EXPOSURE: The patients with keratoconus underwent standard epithelial-off UV-A/riboflavin CXL treatment with 30-minute UV-A exposure at 3 mW/cm2 irradiance. MAIN OUTCOMES AND MEASURES: Central corneal subbasal nerve density and subbasal nerve architecture by use of laser-scanning in vivo confocal microscopy; subbasal nerve analysis by 2 masked observers and by use of a fully automated method; wide-field mosaics of subbasal nerve architecture by use of an automated method; and ocular surface touch sensitivity by use of contact esthesiometry. RESULTS: Mean (SD) age of the 19 patients with keratoconus was 27.5 (7.1) years (range, 19-44 years), and minimal corneal thickness was 428 (36) µm (range, 372-497 µm). Compared with the mean (SD) preoperative subbasal nerve density of 21.0 (4.2) mm/mm2 in healthy corneas, the mean (SD) preoperative subbasal nerve density of 10.3 (5.6) mm/mm2 in the corneas of patients with stage 1 or 2 keratoconus was reduced 51% (mean difference, 10.7 mm/mm2 [95% CI, 6.8-14.6 mm/mm2]; P < .001). After CXL, nerves continued to regenerate for up to 5 years, but nerve density remained reduced relative to healthy corneas at final follow-up (mean reduction, 8.5 mm/mm2 [95% CI, 4.7-12.4 mm/mm2]; P < .001) despite recovery of touch sensitivity to normal levels by 6 months. Preoperatively, more frequent nerve loops, crossings, and greater crossing angles were observed in the corneas of patients with keratoconus compared with healthy corneas. Postoperatively, the frequency of nerve looping increased, crossings were more frequent, and nerve tortuosity increased. Wide-field mosaics indicated persistent disrupted orientation of the regenerating subbasal nerves 5 years after CXL. CONCLUSIONS AND RELEVANCE: Keratoconus is characterized by a neurotrophic deficit and altered nerve morphology that CXL treatment does not address, despite providing a positive biomechanical effect in the stroma. Given the widespread use of CXL in the management of patients with keratoconus, the progression of abnormal innervation after CXL should be recognized.

Focused Tortuosity Definitions Based on Expert Clinical Assessment of Corneal Subbasal Nerves.

PURPOSE: We examined agreement among experts in the assessment of corneal subbasal nerve tortuosity. METHODS: Images of corneal subbasal nerves were obtained from investigators at seven sites (Auckland, Boston, Linköping, Manchester, Oslo, Rostock, and Sydney) using laser-scanning in vivo confocal microscopy. A set of 30 images was assembled and ordered by increasing tortuosity by 10 expert graders from the seven sites. In a first experiment, graders assessed tortuosity without a specific definition and performed grading three times, with at least 1 week between sessions. In a second experiment, graders assessed the same image set using four focused tortuosity definitions. Intersession and intergrader repeatability for the experiments were determined using the Spearman rank correlation. RESULTS: Expert graders without a specific tortuosity definition had high intersession (Spearman correlation coefficient 0.80), but poor intergrader (0.62) repeatability. Specific definitions improved intergrader repeatability to 0.79. In particular, tortuosity defined by frequent small-amplitude directional changes (short range tortuosity) or by infrequent large-amplitude directional changes (long range tortuosity), indicated largely independent measures and resulted in improved repeatability across the graders. A further refinement, grading only the most tortuous nerve in a given image, improved the average correlation of a given grader's ordering of images with the group average to 0.86 to 0.90. CONCLUSIONS: Definitions of tortuosity specifying short or long-range tortuosity and considering only the most tortuous nerve in an image improved the agreement in tortuosity grading among a group of expert observers. These definitions could improve accuracy and consistency in quantifying subbasal nerve tortuosity in clinical studies.

Automatic detection of microdots in the stromal layer of corneal images.

Microdots are bright, 1-2um features of the cornea. It has not been proven what these dots represent, but they are thought to be remnants of apoptotic cell death, such as lipofuscin granules. Their presence has been shown to correlate with corneal aging and extended contact use, both of which are linked to oxygen deprivation in the cornea. Confocal images of the stroma show these microdots mixed with larger keratocyte cells. This paper presents a method for detecting microdots using a two-step filtering scheme that separates the keratocyte cells and the microdots. Keratocyte cell locations are then used to eliminate falsely detected microdots. Results are compared to ground truth based on a grading scale from 0-5. Two graders were given a set of 50 images to grade using a GUI that included sample images for each of the six grades. The two graders had a correlation of .88 with each other. The algorithm had a correlation of .88 with the average of graders and .85 with each of the graders individually.

A fully-automatic fast segmentation of the sub-basal layer nerves in corneal images.

Corneal nerves changes have been linked to damage caused by surgical interventions or prolonged contact lens wear. Furthermore nerve tortuosity has been shown to correlate with the severity of diabetic neuropathy. For these reasons there has been an increasing interest on the analysis of these structures. In this work we propose a novel, robust, and fast fully automatic algorithm capable of tracing the sub-basal plexus nerves from human corneal confocal images. We resort to logGabor filters and support vector machines to trace the corneal nerves. The proposed algorithm traced most of the corneal nerves correctly (sensitivity of 0.88 ± 0.06 and false discovery rate of 0.08 ± 0.06). The displayed performance is comparable to a human grader. We believe that the achieved processing time (0.661 ± 0.07 s) and tracing quality are major advantages for the daily clinical practice.

Automatic montaging of corneal sub-basal nerve images for the composition of a wide-range mosaic.

We present and discuss a computerized system able to provide a wide-range mosaic of the sub-basal nerve layer of central cornea, built from several images acquired in-vivo with confocal microscopy. The montage is performed by a fast, reliable and fully automatic computerized system that does not require any expedient or manual adjustment during the acquisition process. The resulting mosaic provides a large high quality image, which should significantly aid clinicians in evaluating and assessing in a more reliable way the pathologic signs of interest.

Super-image mosaic of infant retinal fundus: selection and registration of the best-quality frames from videos.

Wide-field retinal fundus cameras are commercially available devices that allow acquiring videos of a wide area of infants' eye, considered of clinical interest in screening for ROP (Retinopathy of Prematurity). Many frames of the video are often altered by defects such as artifacts, interlacing and defocus, which make critical and time consuming the search and choice of the good frames to be analyzed. We developed a computerized system that automatically selects the best still frames from the video and builds a mosaic from these images. It will allow clinicians to examine a single large, best quality image. The best frames are identified using several image quality parameters that measure sharpness and steadiness, and then registered to obtain a single mosaic image. A custom blending procedure is then applied in order to provide a final image with homogeneous luminosity and contrast, devoid of the dark areas typically present in the outer regions of single frames. The best-frame selection module showed a PPV of 0.92, while the visual inspection of resulting mosaics confirmed the remarkable capability of the proposed system to provide higher quality images.

Standardized baseline human corneal subbasal nerve density for clinical investigations with laser-scanning in vivo confocal microscopy.

PURPOSE: We established a baseline value for central corneal subbasal nerve density in a large, healthy cohort. METHODS: A total of 106 healthy volunteers (207 eyes) underwent full ophthalmic examination, including laser-scanning in vivo confocal microscopy (IVCM) of the central cornea. Images of the corneal subbasal nerve plexus were acquired and analyzed based on defined criteria. Nerve tracing was performed by two human observers and by a fully automated method. Subbasal nerve density was stratified by eye, observer, tracing method, calculation method, and age group. Association of nerve density with age was examined by linear regression and population distribution was examined by nonlinear regression. RESULTS: We analyzed 892 distinct, high quality images of the subbasal nerve plexus (mean, 4.3 images/eye) from 207 eyes. An overall mean central subbasal nerve density of 19 mm/mm(2) was found in 106 subjects aged 15 to 88 years, independent of eye, sex, or nerve tracing method, while the SD was a consistent 4 to 5 mm/mm(2). Subbasal nerve density followed a normal Gaussian distribution, and correlated negatively with age, with a mean decline of 0.25% to 0.30% per year, independent of eye, observer, or nerve tracing method. Moreover, the use of automated tracing techniques and randomized sampling may improve the speed and reproducibility of subbasal nerve density assessment for clinical applications. CONCLUSIONS: A baseline human corneal subbasal nerve density has been determined by laser-scanning IVCM using rigorous methods. The methods and results could aid in the future assessment of corneal nerves in various patient populations.

Image-level tortuosity estimation in wide-field retinal images from infants with Retinopathy of Prematurity.

Tortuosity and dilation of retinal vessels are considered of primary importance for the diagnosis and follow-up of the Retinopathy of Prematurity (ROP) disease. We developed an algorithm to estimate vessel tortuosity in images acquired with a wide-field fundus camera in ROP subjects, offering clinicians a quantitative, objective, and reproducible diagnostic parameter. Vessels were manually traced in 20 images to provide error-free input data for the tortuosity estimation. At first we investigated different vessel-level measures, some including also caliber information. Then we used them to obtain different imagelevel tortuosity measures, which were eventually combined in a supervised approach to provide a tortuosity index capable to reproduce the clinical experts assessment.To provide manual assessment, the 20 images were independently ordered by increasing tortuosity by three clinical graders and three retinal imaging experts. The proposed tortuosity index obtains a Spearman's correlation coefficient of 0.95 with ground truth, a performance comparable to the clinical graders' one and better than the retinal imaging experts' one.

A modular framework for the automatic classification of chromosomes in Q-band images.

The manual analysis of the karyogram is a complex and time-consuming operation, as it requires meticulous attention to details and well-trained personnel. Routine Q-band laboratory images show chromosomes that are randomly rotated, blurred or corrupted by overlapping and dye stains. We address here the problem of robust automatic classification, which is still an open issue. The proposed method starts with an improved estimation of the chromosome medial axis, along which an established set of features is then extracted. The following novel polarization stage estimates the chromosome orientation and makes this feature set independent on the reading direction along the axis. Feature rescaling and normalizing techniques take full advantage of the results of the polarization step, reducing the intra-class and increasing the inter-class variances. After a standard neural network based classification, a novel class reassignment algorithm is employed to maximize the probability of correct classification, by exploiting the constrained composition of the human karyotype. An average 94% of correct classification was achieved by the proposed method on 5474 chromosomes, whose images were acquired during laboratory routine and comprise karyotypes belonging to slightly different prometaphase stages. In order to provide the scientific community with a public dataset, all the data we used are publicly available for download.

A review of thresholding strategies applied to human chromosome segmentation.

Karyotype analysis is a widespread procedure in cytogenetics to assess the presence of genetic defects by the visualization of the structure of chromosomes. The procedure is lengthy and repetitive and an effective automatic analysis would greatly help the cytogeneticist routine work. Still, automatic segmentation and the full disentangling of chromosomes are open issues. The first step in every automatic procedure is the thresholding step, which detect blobs that represent either single chromosomes or clusters of chromosomes. The better the thresholding step, the easier is the subsequent disentanglement of chromosome clusters into single entities. We implemented eleven thresholding methods, i.e. the ones that appear in the literature as the best performers, and compared their performance in segmenting chromosomes and chromosome clusters in cytogenetic Q-band images. The images are affected by the presence of hyper- or hypo-fluorescent regions and by a contrast variability between the stained chromosomes and the background. A thorough analysis of the results highlights that, although every single algorithm shows peculiar strong/weak points, Adaptive Threshold and Region Based Level Set have the overall best performance. In order to provide the scientific community with a public dataset, the data and manual segmentation used in this paper are available for public download at http://bioimlab.dei.unipd.it.

An improved classification scheme for chromosomes with missing data.

Karyotyping, or the automatic classification of human chromosomes, is mostly based on the analysis of the chromosome specific banding pattern. Unfortunately, the most informative phases of the cell division cycle are composed of long chromosomes that easily overlap: the involved banding pattern information is corrupted, resulting in a drastic increase of the classification error. Assuming the availability of a probabilistic classifier, the improvement of the classification of chromosomes with corrupted data would require the additional estimation of the joint probability density of the observed and missing data for each chromosome class. Given the number of classes, the possible position and extension of the corrupted data within a chromosome, and the dimensionality of the feature space, a reliable estimation would need an impossible number of training samples. We chose to circumvent the estimation problem by developing a statistical generative model of the pattern of each class, so that the corrupted part can be substituted with a partial pattern synthetically generated from the model. This allows to obtain a Monte Carlo estimate of the maximum a posteriori probability for the class given the observation and the missing data, which reduces to a simple voting scheme if the a priori probability for each class is equal. Moreover, this Monte Carlo classification is superior to the voting scheme based on the simple imputation of the classes mean to the missing data.

A Web-based system for the quantitative and reproducible assessment of clinical indexes from the retinal vasculature.

A novel system for the vascular tree identification and the quantitative estimation of arteriolar venular ratio clinical index in retinal fundus images is presented. The system is composed of a module for automatic vascular tracking, an interactive editing interface to correct errors and set the required parameters of analysis, and a module for the computation of clinical indexes. The system was organized as a client-server structure to allow clinicians and researchers from all over the world to work remotely. The system was evaluated by three graders analyzing 30 fundus images. The evaluation of the Pearson's correlation coefficient and p-value of a paired t-test for each pair of graders demonstrates the high reproducibility of the measures provided by the system.

From laboratory to clinic: the development of web-based tools for the estimation of retinal diagnostic parameters.

Over the years, tools for the analysis of retinal images have been developed by several research groups but their usage has been mainly confined within the developing institutions. One possibility to foster their adoption is to develop them as web-based tools. We present here three such systems we recently developed. They are specifically focused on the estimation of retinal vascular parameters, such as arteriolar narrowing (AVR parameter), vessel tortuosity, and vessel caliber narrowing and tortuosity in retinopathy of prematurity (ROP) images. These systems have been successfully evaluated as regards their reliability and will soon be publicly available to interested health care providers.

Automatic vessel segmentation in wide-field retina images of infants with retinopathy of prematurity.

The earliest signs of Retinopathy of Prematurity (ROP) are tortuosity and dilation of retinal vessels. Such vascular changes are considered of primary importance for the diagnosis and the follow-up of the disease. However, a widely accepted computerized system for their quantitative measurement is still missing. Images taken from a preterm baby's eye are often low-contrast, noisy, and blurred. Algorithms that have been successfully applied to analyze adult retinal images do not work well in ROP images. We propose here a novel method for the automatic extraction of vessel centerline in wide-field ROP retinal images, based on a sparse tracking scheme. After a set of seed points is identified all over the image, vessels are traced by connecting those seeds by means of minimum cost paths, whose weights depend on similarity features and alignment evaluated by a custom line operator. The performance of the method was assessed on a dataset of 20 images acquired with the RetCam fundus camera. A sensitivity of 0.78 and a false detection rate of 0.15 were obtained with respect to manual ground truth reference.

Automatic segmentation and disentangling of chromosomes in Q-band prometaphase images.

Karyotype analysis is a widespread procedure in cytogenetics to assess the possible presence of genetics defects. The procedure is lengthy and repetitive, so that an automatic analysis would greatly help the cytogeneticist routine work. Still, automatic segmentation and full disentangling of chromosomes are open issues. We propose an automatic procedure to obtain the separated chromosomes, which are then ready for a subsequent classification step. The segmentation is carried out by means of a space-variant thresholding scheme, which proved to be successful even in presence of hyper- or hypofluorescent regions in the image. Then, the tree of choices to resolve touching and overlapping chromosomes is recursively explored, choosing the best combination of cuts and overlaps based on geometric evidence and image information. We show the effectiveness of the proposed method on routine data acquired with different microscope-camera setup at different laboratories: from 162 images of 117 cells totaling 6683 chromosomes, 94% of the chromosomes were correctly segmented, solving 90% of the overlaps and 90% of the touchings. In order to provide the scientific community with a public dataset, the data used in this paper are available for public download.

Automatic classification of chromosomes in Q-band images.

The manual analysis of the karyogram is a complex, wearing and time-consuming operation. It requires a very meticulous attention to details and calls for well-trained personnel. Even though existing commercial software packages provide a reasonable support to cytogenetists, they very often require human intervention to correct challenging situations. We developed a robust automatic classification system conceived to cope with routine images in which chromosomes are randomly rotated, possibly blurred or also corrupted by overlapping or by dye stains. It consists in a sequence of modules comprising robust feature extraction based on medial axis, chromosome polarization, feature pre-processing, and Neural Network classification followed by a class reassigning algorithm.We show the effectiveness of the proposed method on data comprising karyotypes belonging to slightly different stage of the prometaphase. This dataset contains 119 karyotypes (5474 chromosomes), 70 of which were used for training and validation and 49 for the final testing. In this latter set of images, the system achieved a classification accuracy, as compared to manual ground truth, of 95.6%.

Automatic segmentation of chromosomes in Q-band images.

Karyotype analysis is a widespread procedure in cytogenetics to assess the possible presence of genetics defects. The procedure is lengthy and repetitive, so that an automatic analysis would greatly help the cytogeneticist routine work. Still, automatic segmentation and full disentangling of chromosomes are open issues. We propose an automatic procedure to obtain the separated chromosomes, which are then ready for a subsequent classification step. The segmentation is carried out by means of a space variant thresholding scheme, which proved to be successful even in presence of hyper- or hypo-fluorescent regions in the image. Then a greedy approach is used to identify and resolve touching and overlapping chromosomes, based on geometric evidence and image information. We show the effectiveness of the proposed method on routine data: 90% of the overlaps and 92% of the adjacencies are resolved, resulting in a correct segmentation of 96% of the chromosomes.