Translational Vision Science & Technology最新文献

Purpose: The purpose of this study was to develop and validate a deep-learning model for noninvasive anemia detection, hemoglobin (Hb) level estimation, and identification of anemia-related retinal features using fundus images.

Methods: The dataset included 2265 participants aged 40 years and above from a population-based study in South India. The dataset included ocular and systemic clinical parameters, dilated retinal fundus images, and hematological data such as complete blood counts and Hb concentration levels. Eighty percent of the dataset was used for algorithm development and 20% for validation. A deep-convolutional neural network, utilizing VGG16, ResNet50, and InceptionV3 architectures, was trained to predict anemia and estimate Hb levels. Sensitivity, specificity, and accuracy were calculated, and receiver operating characteristic (ROC) curves were generated for comparison with clinical anemia data. GradCAM saliency maps highlighted regions linked to anemia and image processing techniques to quantify anemia-related features.

Results: For predicting anemia, the InceptionV3 model demonstrated the best performance, achieving 98% accuracy, 99% sensitivity, 97% specificity, and an area under the curve (AUC) of 0.98 (95% confidence interval [CI] = 0.97-0.99). For estimating Hb levels, the mean absolute error for the InceptionV3 model was 0.58 g/dL (95% CI = 0.57-0.59 g/dL). The model focused on the area around the optic disc and the neighboring retinal vessels, revealing that anemic subjects exhibited significantly increased vessel tortuosity and reduced vessel density (P < 0.001), with variable effects on vessel thickness.

Conclusions: The InceptionV3 model accurately predicted anemia and Hb levels, highlighting the potential of deep learning and vessel analysis for noninvasive anemia detection.

Translational relevance: The proposed method offers the possibility to quantitatively predict hematological parameters in a noninvasive manner.

Purpose: Uncorrected visual impairment (VI) significantly impacts life quality and exacerbates age-related health issues. Social determinants of health (SDOH) are associated with uncorrected VI, but quantitative evidence is limited. This study investigated the link between SDOH and uncorrected VI among aging adults to identify disparities and improve vision care.

Methods: We used data from the Atherosclerosis Risk in Communities (ARIC) study visits 4 and 6 and the ancillary Eye Determinants of Cognition (EyeDOC) study. We included subjects who were >70 years old and extracted their sex, race, residence, household income, education level, having an eye doctor, health insurance status, and Area Deprivation Index (ADI) and vision outcomes. Uncorrected VI was categorized into uncorrected distance (UDVI) or near visual impairment (UNVI). Associations between SDOH indicators and VI were evaluated using logistic regressions.

Results: Among 967 adults (mean ± SD age, 78.6 ± 4.35 years; 37.9% male), UDVI was found in 293 and UNVI in 186. Living in Jackson, MS, was associated with lower odds for UNVI (adjusted odds ratio [aOR] = 0.36; 95% CI, 0.20-0.65). Higher odds for UNVI were associated with male sex (aOR = 2.01; 95% CI, 1.41-2.87), low educational attainment (aOR for not completing high school = 2.32; 95% CI, 1.37-3.92; aOR for high school only = 1.92; 95% CI, 1.26-2.92), no eye doctor (aOR = 1.58; 95% CI, 1.05-2.39), and having government health insurance only (aOR = 1.48; 95% CI, 1.00-2.17). Associations between SDOH factors and UDVI were weaker or non-existent.

Conclusions: This study links SDOH factors to uncorrected VI among older adults.

Translational relevance: SDOH should be considered when designing interventions to reduce VI in vulnerable communities.

Purpose: The purpose of this study was to characterize whether pulsed ultrasound (PUS) affects transscleral drug delivery.

Methods: Fluorescein sodium (NaF, 376 Da) and fluorescein isothiocyanate-conjugated dextran 40 (FD-40, 40 kDa) were used as model drugs. Human sclera grafts were placed in modified Franz diffusion cells and were treated by PUS (1 megahertz [MHz], 0.71 W/cm2, duty cycle 30%, application time 5 minutes) once or repeatedly under various conditions to assess permeation enhancement and reservoir effect. The safety of PUS application was assessed on human sclera grafts ex vivo and rabbit eyes in vivo by histology and temperature measurements.

Results: Single PUS application yielded a significant increase in FD-40 permeation (P <  0.05). Repeated PUS applications led to a further enhancement in FD-40 permeation and also significantly promoted NaF permeation (more than 8.51-fold, P <  0.05). The human scleral permeability was temporarily modified by PUS, as evidenced by the increased scleral permeability during PUS application and the unchanged permeability coefficients at steady state. The reservoir effect of human sclera was also enhanced by PUS application. Cavitation was detected under PUS. A minor increase in graft temperature rise (<1°C) and no ocular damage was caused by PUS.

Conclusions: PUS is an efficient and safe method to enhance model drugs to transport across human sclera by increasing the scleral permeability transiently and improving the reservoir effect. The enhancement was correlated with the molecule size and further promoted by the repeated PUS application.

Translational relevance: Our study provides proof of concept for using PUS to enhance drug delivery to the posterior eye segment.

Purpose: To compare a novel high-resolution optical coherence tomography (OCT) with improved axial resolution (High-Res OCT) with conventional spectral-domain OCT (SD-OCT) with regard to their capacity to characterize the disorganization of the retinal inner layers (DRIL) in diabetic maculopathy.

Methods: Diabetic patients underwent multimodal retinal imaging (SD-OCT, High-Res OCT, and color fundus photography). Best-corrected visual acuity and diabetes characteristics were recorded. DR was graded using the international clinical diabetic retinopathy severity scale (DRSS). In each OCT B-scan, retinal layers were segmented and the loss of discernibility was annotated. DRIL areas were analyzed in en face projection using FIJI plugins. The Wilcoxon test and regression models were used for statistical analysis.

Results: In 93 eyes of 93 patients (mean age, 61.8 ± 12.9 years) DRIL was identified in 48 eyes. DRIL was most frequent in the central subfield (27%). In DRIL eyes, DRSS was significantly higher (4.43 ± 1.01 vs. 2.12 ± 1.66; P < 0.001), BCVA was significantly worse (0.34 ± 0.38 vs. 0.13 ± 0.22; P < 0.001), and the loss of discernibility of the individual inner retinal layers was significantly smaller in High-Res OCT compared with SD-OCT (0.21 ± 0.29 vs. 1.21 ± 1.21 mm2; P < 0.001). The discernibility loss was greatest in the retinal nerve fiber layer and ganglion cell layer.

Conclusions: DRIL occurs in eyes with advanced diabetic retinopathy, with a characteristic spread: from the inner toward the outer retina. High-Res OCT shows significantly smaller DRIL areas compared with SD-OCT, because of a more precise delineation of the inner retinal layers.

Translational relevance: Using OCT with increased axial resolution could enhance our understanding of DRIL development and progression, providing deeper insights into pathophysiological aspects, including malperfusion in the inner capillary plexus.

Purpose: Individuals with Down syndrome (DS) have reduced visual acuity (VA), even when wearing refractive correction. The relationship between refractive error and VA in adults with DS is explored.

Methods: Thirty adults with DS (age = 29 ± 10 years) were enrolled in a trial comparing clinical and objectively determined refractions. Monocular VA was recorded unaided and aided with best refraction. Vectors M, J0, and J45 were calculated from unaided wavefront aberration measures at the habitual pupil size. The square root of the sum of the squared vectors was calculated providing a single positive vector length representing unaided refractive error. Residual refractive error was determined after applying the best performing refraction. Linear regression determined correlation between refractive error and VAs.

Results: Unaided and aided VAs ranged from 0.22 to 1.42 logMAR and 0.06 to 0.82 logMAR, respectively. Unaided and residual refractive error represented as vector length ranged from 0.68 diopters (D) to 13.76 D and 0.05 D to 1.87 D, respectively. Unaided refractive error and VA were significantly positively correlated (r2 = 0.776, P < 0.001), but not residual refractive error and VA (r2 = 0.005, P = 0.721).

Conclusions: There was a positive correlation between unaided VA and refractive error magnitude in adults with DS; however, unaided VA was better than expected given the high levels of refractive error. Aided VA and residual refractive error were not correlated, despite overall low levels of remaining residual refractive error, suggesting that factors in addition to optical quality may be limiting VA in this population.

Translational relevance: Understanding the relationship between refractive error and VA in individuals with DS may provide clinicians clearer expectations for the acuity end points before and after correction for this patient population.

Purpose: To extract conjunctival bulbar redness from standardized high-resolution ocular surface photographs of a novel imaging system by implementing an image analysis pipeline.

Methods: Data from two trials (healthy; outgoing ophthalmic clinic) were collected, processed, and used to train a machine learning model for ocular surface segmentation. Various regions of interest were defined to globally and locally extract a redness biomarker based on color intensity. The image-based redness scores were correlated to clinical gradings (Efron) for validation.

Results: The model to determine the regions of interest was verified for a segmentation performance, yielding mean intersections over union of 0.9639 (iris) and 0.9731 (ocular surface). All trial data were analyzed and a digital grading scale for the novel imaging system was established. Photographs and redness scores from visits weeks apart showed good feasibility and reproducibility. For scores within the same session, a mean coefficient of variation of 4.09% was observed. A moderate positive Spearman correlation (0.599) was found with clinical grading.

Conclusions: The proposed conjunctival bulbar redness extraction pipeline demonstrates that by using standardized imaging, a segmentation model and image-based redness scores' external eye photography can be classified and evaluated. Therefore, it shows the potential to provide eye care professionals with an objective tool to grade ocular redness and facilitate clinical decision-making in a high-throughput manner.

Translational relevance: To empower clinicians and researchers with a high-throughput workflow by standardized imaging combined with an analysis tool based on artificial intelligence to objectively determine an image-based redness score.

Purpose: Previous researches have suggested an important association between gut microbiota (GM) and vascular pathologies such as atherosclerosis. This study aimed to explore the association between 196 GM taxa and retinal vein occlusion (RVO).

Methods: This study used Mendelian randomization (MR), linkage disequilibrium score regression (LDSC), and polygenic overlap analysis. Genome-wide association study (GWAS) data associated with 196 GM taxa was obtained from the MiBioGen consortium, involving a large number of European-ancestry participants. GWAS data of RVO was obtained from the FinnGen consortium and another study that also involved European-ancestry participants. Inverse-variance weighted was used as the primary approach for MR estimation. Moreover, LDSC and polygenic overlap analyses were performed to evaluate the genetic correlation between GM taxa and RVO.

Results: The MR results identified the association of six GM taxa, including class Bacilli, order Lactobacillales, family Streptococcaceae, genus Clostridium innocuum group, genus Family XIII AD3011 group, and genus Subdoligranulum with the development of RVO. In addition, the polygenic overlap analysis supported the genetic association between GM and RVO.

Conclusions: Our findings confirmed the association between six GM taxa and the development of RVO, thereby highlighting the effects of GM on retinal vascular health.

Translational relevance: The results may provide the rationale for developing GM-based strategies for preventing the onset of RVO.

Purpose: Geographic atrophy (GA), an advanced form of dry age-related macular degeneration (AMD), has limited treatment options. This study introduces a novel mouse model featuring an expanding GA patch that can be used to test mechanisms and therapeutics.

Methods: C57Bl/6J male mice (n = 96) aged 9-10 weeks received an intraperitoneal (IP) injection of 20 mg/kg sodium iodate (NaIO3). In vivo confocal scanning laser ophthalmoscope (cSLO) and optical coherence tomography imaging were done at one, four, eight, and 16 weeks after injection, with GA area measurements taken at weeks 8 and 16. Mice were euthanized on weeks 8 and 16 for histological analysis.

Results: Administration of 20 mg/kg intraperitoneal NaIO3 caused variable damage levels. Approximately 22% of cases showed damage (speckled autofluorescence) covering 35% to 90% of the 102° field of view cSLO image at one week after injection. These mice developed an expanding patch of GA by week 8, with a mean 1.45-fold increase in area by week 16. This region showed complete photoreceptor and retinal pigment epithelium loss and complement activation at the atrophy edge, whereas the inner retina remained undamaged. Mice with less damage (48% of cases) only developed incomplete outer retinal degeneration, and mice with more damage (30% of cases) had too much GA for measurable expansion.

Conclusions: Although expanding GA formed in only 22% of mice, the model's simplicity and predictability for GA development via one-week post-injection imaging make it suitable for GA therapeutic experimentation.

Translational relevance: This murine model provides a valuable tool for testing GA therapies, mirroring clinical endpoints relevant to human trials.