Use of explainable AI on slit-lamp images of anterior surface of eyes to diagnose allergic conjunctival diseases.

BACKGROUND: Artificial intelligence (AI) is a promising new technology that has the potential of diagnosing allergic conjunctival diseases (ACDs). However, its development is slowed by the absence of a tailored image database and explainable AI models. Thus, the purpose of this study was to develop an explainable AI model that can not only diagnose ACDs but also present the basis for the diagnosis. METHODS: A dataset of 4942 slit-lamp images from 10 ophthalmological institutions across Japan were used as the image database. A sequential pipeline of segmentation AI was constructed to identify 12 clinical findings in 1038 images of seasonal and perennial allergic conjunctivitis (AC), atopic keratoconjunctivitis (AKC), vernal keratoconjunctivitis (VKC), giant papillary conjunctivitis (GPC), and normal subjects. The performance of the pipeline was evaluated by determining its ability to obtain explainable results through the extraction of the findings. Its diagnostic accuracy was determined for 4 severity-based diagnosis classification of AC, AKC/VKC, GPC, and normal. RESULTS: Segmentation AI pipeline efficiently extracted crucial ACD indicators including conjunctival hyperemia, giant papillae, and shield ulcer, and offered interpretable insights. The AI pipeline diagnosis had a high diagnostic accuracy of 86.2%, and that of the board-certified ophthalmologists was 60.0%. The pipeline had a high classification performance, and the area under the curve (AUC) was 0.959 for AC, 0.905 for normal subjects, 0.847 for GPC, 0.829 for VKC, and 0.790 for AKC. CONCLUSIONS: An explainable AI model created by a comprehensive image database can be used for diagnosing ACDs with high degree of accuracy.

Biologics for allergy: therapeutic potential for ocular allergic diseases and adverse effects on the eye.

Biologics applying antibodies against IgE, IL-5, IL-5 receptor α, IL-4 receptor α, and IL-13 have dramatically improved recent treatment outcomes in allergic diseases including asthma, rhinitis, and atopic dermatitis. However, these drugs have not been approved for ocular allergic diseases such as allergic conjunctivitis, vernal keratoconjunctivitis, and atopic keratoconjunctivitis. Although the putative mechanisms suggest that these drugs should have beneficial effects in patients with ocular allergies and some studies have reported such beneficial effects, various adverse ocular symptoms have also been observed in clinical trials and off-label use studies. Since ocular allergic diseases have distinct pathogeneses, each biologic drug must be examined regarding specific effects on each ocular allergy. For example, IgE-mediated type 1 hypersensitivity plays a critical role in allergic conjunctivitis. By contrast, T cells and eosinophilic and non-IgE-mediated type 2 inflammation play important roles in vernal keratoconjunctivitis. Allergists must fully understand the effects of each drug on the eye. This review outlines both potential therapeutic and adverse effects of various biologics on allergic diseases of the eye.

Role of Damage-Associated Molecular Patterns (DAMPs/Alarmins) in Severe Ocular Allergic Diseases.

Severe ocular allergic diseases, such as atopic keratoconjunctivitis and vernal keratoconjunctivitis, cause severe allergic inflammation in the conjunctiva and corneal epithelial damage, resulting in visual disturbances. The involvement of damage (danger)-associated molecular patterns (DAMPs/alarmins) in the pathogenesis of these diseases has been recognized. Alarmins released from damaged corneal epithelial cells or eosinophils play a critical role in the induction of corneal lesions, vicious loop of corneal injury, and exacerbation of conjunctival allergic inflammation. Alarmins in the conjunctiva also play an essential role in the development of both allergic inflammation, based on the acquired immune system, and type 2 inflammation by innate immune responses in the ocular surface. Therefore, alarmins may be a potentially important therapeutic target in severe refractory ocular allergic diseases.

Severity Classification of Conjunctival Hyperaemia by Deep Neural Network Ensembles.

Conjunctival hyperaemia is a common clinical ophthalmological finding and can be a symptom of various ocular disorders. Although several severity classification criteria have been proposed, none include objective severity criteria. Neural networks and deep learning have been utilised in ophthalmology, but not for the purpose of classifying the severity of conjunctival hyperaemia objectively. To develop a conjunctival hyperaemia grading software, we used 3700 images as the training data and 923 images as the validation test data. We trained the nine neural network models and validated the performance of these networks. We finally chose the best combination of these networks. The DenseNet201 model was the best individual model. The combination of the DenseNet201, DenseNet121, VGG19, and ResNet50 were the best model. The correlation between the multimodel responses, and the vessel-area occupied was 0.737 (p < 0.01). This system could be as accurate and comprehensive as specialists but would be significantly faster and consistent with objective values.

Evaluation of offset of conjunctival hyperemia induced by a Rho-kinase inhibitor; 0.4% Ripasudil ophthalmic solution clinical trial.

Glaucoma leads to irreversible blindness. Numerous anti-glaucoma eye drops have been developed. Unfortunately, many patients with glaucoma still suffer from progressive visual disorders. Recently, ripasudil hydrochloride hydrate, a selective Rho-associated protein kinase inhibitor, was launched for the treatment of glaucoma. However, adverse events, such as conjunctival hyperemia, are often noted in clinical trials using healthy subjects. Therefore, we investigated the onset, offset, and kinetic changes of conjunctival hyperemia induced by ripasudil ophthalmic solution in patients with open-angle glaucoma or ocular hypertension who had already been treated with anti-glaucoma eye drops other than ripasudil. Conjunctival hyperemia was evaluated by both clinical grading by 3 ophthalmic physicians and pixel coverage of conjunctival blood vessels determined by conjunctival hyperemia-analyzing software. Conjunctival hyperemia appeared within 10 min post-instillation in most of the participants. Clinical grade and pixel coverage increased significantly 10 min post-instillation and then decreased. In most of the participants, hyperemia resolved within 2 h. Median conjunctival hyperemia offset was 90 min. A tendency of monotonic increase was observed between clinical grade and pixel coverage. Taken altogether, hyperemia induced by ripasudil was transient in glaucoma patients who had already been treated with anti-glaucoma eye drops other than ripasudil.

Hyperemia Analysis Software for Assessment of Conjunctival Hyperemia Severity.

PURPOSE: We developed a hyperemia analysis software, which can quantitatively assess the degree of conjunctival hyperemia, and evaluated the reproducibility and reliability of its percent coverage. In addition, we compared the clinical grading and the percent coverage to examine the applicability of the software analysis. METHODS: We took images of the temporal conjunctiva with slit lamp microscopes. We used our hyperemia analysis software to detect blood vessels in the region of interest using its image processing capabilities and obtained a ratio of the area occupied by blood vessels to the rest of the area. In addition, we used the clinical evaluation criteria of the Japanese guidelines for allergic conjunctival diseases 2017 to clinically grade the hyperemia in each eye. Nine evaluators examined the images and graded the severity of hyperemia into four stages. We looked for a statistical correlation between the results of the hyperemia analysis software and the clinical grading. RESULTS: The percent coverage of the blood vessels in the region of interest calculated by the hyperemia analysis software correlated with the arithmetic average of our clinical grading (r = 0.953; 95% CI, 0.8470340-0.9862136). CONCLUSION: The percent coverage from our hyperemia analysis software reflects the clinical grading score, suggesting that our software can be used to obtain a detailed analysis of conjunctival hyperemia.

Requirement of longer term antiviral therapy in patients with cytomegalovirus anterior uveitis with corneal endothelial cell damage.

BACKGROUND: The aim of the study was to investigate the efficacy of therapy in patients with cytomegalovirus (CMV) anterior uveitis. PATIENTS AND METHODS: We reviewed the records of patients with CMV anterior uveitis who attended our institution between October 2010 and December 2015 and who were confirmed to have CMV DNA in the aqueous humor by polymerase chain reaction analysis. RESULTS: Fourteen immunocompetent patients (10 men and 4 women, total of 17 eyes) were enrolled. The mean ± SD age at the onset of antiviral therapy was 63.1 ± 11.3 years (range, 44-87 years). CMV DNA was not detected in the aqueous humor of 3 patients on initial testing, but it was detected on subsequent analysis. All patients underwent systemic antiviral therapy. Among the patients who were followed up for more than 6 months after starting systemic antiviral therapy, systemic antiviral therapy was successfully terminated in all 4 patients without corneal endothelial loss but had to be continued because of disease recurrence on its termination in 5 of 8 patients (62.5%) with corneal endothelial damage (P = 0.038). CONCLUSIONS: Patients with corneal endothelial cell loss are likely to require longer term antiviral therapy than those without endothelial damage. In addition, whereas definitive diagnosis of CMV anterior uveitis requires the detection of CMV DNA in aqueous humor by polymerase chain reaction, one-fifth of patients in the present study tested negative on initial examination.

Clinical Severity Classification using Automated Conjunctival Hyperemia Analysis Software in Patients with Superior Limbic Keratoconjunctivitis.

PURPOSE: Digitization of clinical observation is necessary for assessing the severity of superior limbic keratoconjunctivitis (SLK). This study aimed to use a novel quantitative marker to examine hyperemia in patients with SLK. MATERIALS AND METHODS: We included six eyes of six patients with both dry eye disease and SLK (SLK group) and eight eyes of eight patients with Sjögren syndrome (SS group). We simultaneously obtained the objective finding scores by using slit-lamp examination and calculated the superior hyperemia index (SHI) with an automated conjunctival hyperemia analysis software by using photographs of the anterior segment. Three objective finding scores, including papillary formation of the superior palpebral conjunctiva, superior limbal hyperemia and swelling, and superior corneal epitheliopathy, were determined. The SHI was calculated as the superior/temporal ratio of bulbar conjunctival hyperemia by using the software. Fisher's exact test was used to compare a high SHI (≥1.07) ratio between the SLK and SS groups. P-Values < 0.05 were considered statistically significant. RESULTS: The SHI (mean ± standard deviation) in the SLK and SS groups was 1.19 ± 0.50 and 0.69 ± 0.24, respectively. The number of patients with a high SHI (≥1.07) was significantly higher in the SLK group than in the SS group (p < 0.05). The sensitivity and specificity of the SHI in the differential diagnosis between SS and SLK were 66.7% and 87.5%, respectively. An analysis of the association between the objective finding scores and SHI showed that the SHI had a tendency to indicate the severity of superior limbal hyperemia and swelling score in the SLK group. CONCLUSION: The SHI calculated using the automated conjunctival hyperemia analysis software could successfully quantify superior bulbar conjunctival hyperemia and may be a useful tool for the differential diagnosis between SS and SLK and for the quantitative follow-up of patients with SLK.

Association between glaucoma eye drops and hyperemia.

PURPOSE: We used an image analysis software program to quantitatively investigate conjunctival injection in patients treated with eye drops for glaucoma. METHODS: We compared 169 patients (89 men and 80 women) with a diagnosis of glaucoma. Photographs of the conjunctiva were taken on the temporal side of each patient's right eye using a slit lamp. We determined the mean pixel frequencies of the conjunctival blood vessels from the photographs. RESULTS: The ocular hyperemia of the patients being treated with prostaglandins was more severe than that of the patients being treated with beta-blockers or no eye drops. In multiple comparisons of each of the eye drops, the control group had a significantly lower degree of hyperemia than did the patients being treated with each of the various prostaglandin analogs (latanoprost, travoprost, tafluprost, and bimatoprost). Among the patients receiving prostaglandin, the percentage of those with hyperemia was highest in the bimatoprost users, followed in order by the travoprost, latanoprost, and tafluprost users. However, no significant differences were found among the different prostaglandin analogs in terms of the percentage of patients with hyperemia. CONCLUSION: Our software program may be useful for evaluating the hyperemic effects of eye drops used for glaucoma. The particular type of prostaglandin analog seems to determine the level of conjunctival hyperemia during ocular hypotensive medical treatment.

Development of automated conjunctival hyperemia analysis software.

Conjunctival hyperemia is observed in a variety of ocular inflammatory conditions. The evaluation of hyperemia is indispensable for the treatment of patients with ocular inflammation. However, the major methods currently available for evaluation are based on nonquantitative and subjective methods. Therefore, we developed novel software to evaluate bulbar hyperemia quantitatively and objectively. First, we investigated whether the histamine-induced hyperemia of guinea pigs could be quantified by image analysis. Bulbar conjunctival images were taken by means of a digital camera, followed by the binarization of the images and the selection of regions of interest (ROIs) for evaluation. The ROIs were evaluated by counting the number of absolute pixel values. Pixel values peaked significantly 1 minute after histamine challenge was performed and were still increased after 5 minutes. Second, we applied the same method to antigen (ovalbumin)-induced hyperemia of sensitized guinea pigs, acquiring similar results except for the substantial upregulation in the first 5 minutes after challenge. Finally, we analyzed human bulbar hyperemia using the new software we developed especially for human usage. The new software allows the automatic calculation of pixel values once the ROIs have been selected. In our clinical trials, the percentage of blood vessel coverage of ROIs was significantly higher in the images of hyperemia caused by allergic conjunctival diseases and hyperemia induced by Bimatoprost, compared with those of healthy volunteers. We propose that this newly developed automated hyperemia analysis software will be an objective clinical tool for the evaluation of ocular hyperemia.

Rebamipide increases barrier function and attenuates TNFα-induced barrier disruption and cytokine expression in human corneal epithelial cells.

BACKGROUND: Disruption of corneal epithelial barrier function by inflammation may contribute to the development of dry eye. The effects of rebamipide, a drug used for the treatment of dry eye, on barrier function and cytokine expression in a human corneal epithelial (HCE) cell line were examined. METHODS: Barrier function of HCE cells was evaluated by measurement of transepithelial electrical resistance (TER). The subcellular localisation of the tight junction protein zonula occludens (ZO)-1 was examined by immunofluorescence analysis. The release of cytokines was determined with ELISAs, and the intracellular abundance of cytokine mRNAs was quantitated by reverse transcription and real-time PCR analysis. Degradation of the nuclear factor-κB inhibitor IκBα was detected by immunoblot analysis. RESULTS: Rebamipide increased TER of HCE cells in a concentration-dependent manner as well as attenuating the loss of TER and the disappearance of ZO-1 from the cell surface induced by tumour necrosis factor α (TNFα). Rebamipide also suppressed TNFα-induced expression of interleukin-6 and interleukin-8 at the mRNA and protein levels and inhibited the TNFα-induced degradation of IκBα. CONCLUSIONS: The upregulation of barrier function and the anti-inflammatory effects of rebamipide, together with its mucin secretagogue activity, may contribute to the effectiveness of this drug for the treatment of dry eye.

Identification of keratocyte-like cells differentiated from circulating bone marrow-derived cells in the mouse cornea.

Bone marrow (BM)-derived stem cells have the potential to differentiate into multiple lineages of tissue resident cells. BM-derived cells have been detected in the mouse cornea, but most of these cells were found to be CD45(+) or CD11b(+) immunocompetent cells. Although some BM-derived cells in the cornea were negative for these cell surface markers, it has remained unclear whether cells of BM origin can differentiate into corneal resident cells. To address this issue, we subjected wild-type mice that had been exposed to a lethal dose of radiation to intravenous injection with BM cells from green fluorescent protein (GFP) transgenic mice. Two months after cell transplantation, fluorescence microscopy revealed the presence of numerous GFP(+) cells throughout the cornea, with intense GFP fluorescence being apparent around the limbal region. Immunohistofluorescence analysis of corneal cross-sections revealed that most of the BM-derived GFP(+) cells expressed CD45 or CD11b, although a few GFP(+) cells were negative for these markers. Immunostaining of individual cells isolated from the corneal stroma, however, showed that a small proportion (~1 %) of GFP(+) BM-derived cells expressed the keratocyte-specific proteoglycan keratocan. Our results suggest that BM-derived cells introduced intravenously are able to differentiate into resident cells of the corneal stroma.

Oral administration of Ag suppresses Ag-induced allergic conjunctivitis in mice: critical timing and dose of Ag.

BACKGROUND/AIMS: To investigate the effect of Ag doses and administration time points on oral tolerance induction in experimental allergic conjunctivitis (EAC). METHODS: BALB/c mice were actively sensitised twice with ovalbumin (OVA) in alum, and then challenged twice with OVA in eye drops. Twenty minutes after the last challenge, the clinical appearance was evaluated. Twenty-four hours later, the conjunctivas, spleens and blood were collected for histological analyses, cytokine production assays, and measurement of serum Ig levels, respectively. The mice were fed with a high or low dose of OVA before sensitisation (prophylactic treatment). To assess the effect of therapeutic treatment, the high-dose Ag was administered after sensitisation. Control groups received phosphate-buffered saline without OVA. To determine the role of interleukin-10 (IL-10) and transforming growth factor-ß (TGF-ß) in prophylactic low-dose oral tolerance, mice were injected intraperitoneally with neutralising antibodies during the entire experimental period. To assess the role of regulatory T cells, neonatally thymectomised mice were injected with depleting anti-CD25 monoclonal antibodies before the low-dose prophylactic treatment. RESULTS: OVA-feeding was suppressive even when the mice were treated soon after Ag sensitisation. Both the low and high doses of oral OVA suppressed EAC. High-dose treatment significantly suppressed EAC-related cytokine production and serum Ig levels. IL-10 and TGF-ß were less likely to be involved in prophylactic low-dose oral tolerance induction but regulatory T cells played a role. CONCLUSIONS: Prophylactic and early therapeutic treatment of OVA feeding suppressed EAC. Both high and low doses of oral OVA induced oral tolerance but with different mechanisms.

Pseudomonas aeruginosa keratitis in mice: effects of topical bacteriophage KPP12 administration.

The therapeutic effects of bacteriophage (phage) KPP12 in Pseudomonas aeruginosa keratitis were investigated in mice. Morphological analysis showed that phage KPP12 is a member of the family Myoviridae, morphotype A1, and DNA sequence analysis revealed that phage KPP12 is similar to PB1-like viruses. Analysis of the phage KPP12 genome did not identify any genes related to drug resistance, pathogenicity or lysogenicity, and so phage KPP12 may be a good candidate for therapeutic. KPP12 showed a broad host range for P. aeruginosa strains isolated from clinical ophthalmic infections. Inoculation of the scarified cornea with P. aeruginosa caused severe keratitis and eventual corneal perforation. Subsequent single-dose administration of KPP12 eye-drops significantly improved disease outcome, and preserved the structural integrity and transparency of the infected cornea. KPP12 treatment resulted in the suppression of neutrophil infiltration and greatly enhanced bacterial clearance in the infected cornea. These results indicate that bacteriophage eye-drops may be a novel adjunctive or alternative therapeutic agent for the treatment of infectious keratitis secondary to antibiotic-resistant bacteria.

[A quality of life questionnaire for Japanese allergic conjunctival disease].

PURPOSE: To establish a specific quality of life (QOL) questionnaire for Japanese allergic conjunctival disease (ACD) (Japanese allergic conjunctival disease QOL questionnaire: JACQLQ). SUBJECTS AND METHODS: A multicenter study was conducted in 521 patients with ACD and 127 healthy volunteers (total 648 cases). The JACQLQ ver. 0 was developed by modifying the Japanese rhino-conjunctivitis QOL questionnaire (JRQLQ). The participants were asked to complete the questionnaire, and objective scores were determined by an ophthalmologist using a slit lamp. To confirm validity, item and factor analyses were conducted and correlation coefficients were calculated. RESULTS: The items were grouped into four subscales (Daily activity, Psychological well-being, Eye symptoms, Nasal symptoms) after factor analysis. The JACQLQ had good item-internal consistency (Cronbach's alpha: 0.846-0.934). QOL scores were correlated with eye itching, eye irritation and tearing. Objective scores were correlated with eye redness, eye itching and eye irritation. Face scores were correlated with eye itching, eye irritation and eye redness. CONCLUSION: The JACQLQ is a useful tool for assessing disease specific QOL in ACD.

Automated hyperemia analysis software: reliability and reproducibility in healthy subjects.

PURPOSE: To evaluate the reliability and reproducibility of automated software to analyze human bulbar hyperemia. METHODS: We enrolled 89 healthy volunteers in this study. A slit lamp was used to take pictures of the conjunctiva on the temporal side of each subject's right eye. Photographic conditions were standardized by using a single photographer. Images were transferred to software for automatic pixel value calculation in the green channel of the region of interest (ROI). We investigated optimal ROI size, mean ROI pixel frequency, percentage ROI blood vessel coverage, and data reproducibility. We also used this software to evaluate bimatoprost-induced hyperemia and hyperemia in allergic conjunctival diseases. RESULTS: The optimal ROI was found to be 400 vertical pixels by 300 horizontal pixels. Mean ROI pixel frequency was 5305 and % coverage was 4.4%. We confirmed the reproducibility of the analysis by comparing two images (r (2) = 0.7, P < 0.0001). Percentage blood vessel coverage increased in images of bimatoprost-induced hyperemia and hyperemia in allergic conjunctival diseases compared to the data from healthy volunteers. CONCLUSIONS: The software was simple to use and provided reproducible data. We established standard settings for the operation of the software. The use of our software will improve hyperemia evaluation, which is presently done using nonquantitative methods.

B and T lymphocyte attenuator regulates the development of antigen-induced experimental conjunctivitis.

PURPOSE: To investigate the roles that B and T lymphocyte attenuator (BTLA) and herpesvirus entry mediator (HVEM) play in the development of antigen-induced experimental conjunctivitis (EC). METHODS: BALB/c mice were immunized with ragweed (RW) in alum. Ten days later, the mice were challenged with RW in eye drops. After 24 hours, the conjunctivas, blood and spleens were collected for histological analysis, measurement of serum immunoglobulin (Ig) levels, and both flow cytometric analysis and cytokine assays, respectively. The mice were injected intraperitoneally with anti-BTLA antibody, anti-HVEM antibody or control antibody during either induction phase or effector phase. RESULTS: Induction-phase treatment with anti-BTLA antibody but not anti-HVEM antibody significantly increased conjunctival eosinophil infiltration. Treatment with either antibody during the effector phase did not affect conjunctival eosinophil infiltration. Anti-BTLA antibody treatment during the induction phase reduced the B cell compartment and increased the CD11b-positive cell compartment in splenocytes. Additionally, anti-BTLA treatment upregulated IL-4 and IL-10 production of splenocytes stimulated by RW. CONCLUSIONS: BTLA regulated the development of EC possibly by downregulating Th2 cytokine production and adjusting the compartments of immunocompetent cells. The regulation of EC by BTLA may be mediated by BTLA ligands other than HVEM.

Adjuvants determine the contribution of basophils to antigen sensitization in vivo.

PURPOSE: Basophils expressing FcɛRIα, the alpha chain of the total IgE high affinity receptor, appear to play a role in antigen (Ag) sensitization, although dendritic cells (DCs) are the principal Ag-presenting cells (APCs). To investigate whether these two types of APCs are involved differentially in Ag-sensitization when distinct types of adjuvants are utilized. METHODS: To investigate whether basophils and DCs serve as APCs in vitro, whole splenocytes, FcɛRIα(+) basophil-depleted splenocytes, and CD11c(+) DC-depleted splenocytes from naïve DO11.10 mice were stimulated in vitro with ovalbumin (OVA) or OVA peptide 323-339 to evaluate Ag-induced proliferation. To investigate whether basophils function as APCs in vivo, BALB/c mice were actively immunized with ragweed (RW) emulsified in alum or Complete Freund's Adjuvant (CFA) followed by an RW challenge in eye drops. An anti-FcɛRIα antibody (Ab) or a control Ab was injected intraperitoneally during the sensitization phase. Twenty-four hours after the RW challenge, conjunctivas and spleens were harvested for histological analysis to evaluate conjunctival eosinophilia and cytokine production, respectively. RESULTS: Depletion of basophils or DCs from naïve DO11.10 splenocytes significantly suppressed proliferative responses to either OVA or OVA peptide. Treatment with the anti-FcɛRIα Ab suppressed the conjunctival eosinophilia when alum, but not CFA, was utilized as the adjuvant. Similarly, the anti-FcɛRIα Ab inhibited cytokine production by splenocytes when alum was used as the adjuvant. CONCLUSION: Adjuvants determine which APCs are utilized in Ag-sensitization.