Artificial intelligence-based prediction of diabetic retinopathy evolution (EviRed): protocol for a prospective cohort.

INTRODUCTION: An important obstacle in the fight against diabetic retinopathy (DR) is the use of a classification system based on old imaging techniques and insufficient data to accurately predict its evolution. New imaging techniques generate new valuable data, but we lack an adapted classification based on these data. The main objective of the Evaluation Intelligente de la Rétinopathie Diabétique, Intelligent evaluation of DR (EviRed) project is to develop and validate a system assisting the ophthalmologist in decision-making during DR follow-up by improving the prediction of its evolution. METHODS AND ANALYSIS: A cohort of up to 5000 patients with diabetes will be recruited from 18 diabetology departments and 14 ophthalmology departments, in public or private hospitals in France and followed for an average of 2 years. Each year, systemic health data as well as ophthalmological data will be collected. Both eyes will be imaged by using different imaging modalities including widefield photography, optical coherence tomography (OCT) and OCT-angiography. The EviRed cohort will be divided into two groups: one group will be randomly selected in each stratum during the inclusion period to be representative of the general diabetic population. Their data will be used for validating the algorithms (validation cohort). The data for the remaining patients (training cohort) will be used to train the algorithms. ETHICS AND DISSEMINATION: The study protocol was approved by the French South-West and Overseas Ethics Committee 4 on 28 August 2020 (CPP2020-07-060b/2020-A01725-34/20.06.16.41433). Prior to the start of the study, each patient will provide a written informed consent documenting his or her agreement to participate in the clinical trial. Results of this research will be disseminated in peer-reviewed publications and conference presentations. The database will also be available for further study or development that could benefit patients. TRIAL REGISTRATION NUMBER: NCT04624737.

Impact of the H274Y Substitution on N1, N4, N5, and N8 Neuraminidase Enzymatic Properties and Expression in Reverse Genetic Influenza A Viruses.

The H274Y substitution (N2 numbering) in neuraminidase (NA) N1 confers oseltamivir resistance to A(H1N1) influenza viruses. This resistance has been associated with reduced N1 expression using transfected cells, but the effect of this substitution on the enzymatic properties and on the expression of other group-1-NA subtypes is unknown. The aim of the present study was to evaluate the antiviral resistance, enzymatic properties, and expression of wild-type (WT) and H274Y-substituted NA for each group-1-NA. To this end, viruses with WT or H274Y-substituted NA (N1pdm09 or avian N4, N5 or N8) were generated by reverse genetics, and for each reverse-genetic virus, antiviral susceptibility, NA affinity (Km), and maximum velocity (Vm) were measured. The enzymatic properties were coupled with NA quantification on concentrated reverse genetic viruses using mass spectrometry. The H274Y-NA substitution resulted in highly reduced inhibition by oseltamivir and normal inhibition by zanamivir and laninamivir. This resistance was associated with a reduced affinity for MUNANA substrate and a conserved Vm in all viruses. NA quantification was not significantly different between viruses carrying WT or H274Y-N1, N4 or N8, but was lower for viruses carrying H274Y-N5 compared to those carrying a WT-N5. In conclusion, the H274Y-NA substitution of different group-1-NAs systematically reduced their affinity for MUNANA substrate without a significant impact on NA Vm. The impact of the H274Y-NA substitution on viral NA expression was different according to the studied NA.

Federated Learning for Diabetic Retinopathy Detection in a Multi-center Fundus Screening Network.

Federated learning (FL) is a machine learning framework that allows remote clients to collaboratively learn a global model while keeping their training data localized. It has emerged as an effective tool to solve the problem of data privacy protection. In particular, in the medical field, it is gaining relevance for achieving collaborative learning while protecting sensitive data. In this work, we demonstrate the feasibility of FL in the development of a deep learning model for screening diabetic retinopathy (DR) in fundus photographs. To this end, we conduct a simulated FL framework using nearly 700,000 fundus photographs collected from OPHDIAT, a French multi-center screening network for detecting DR. We develop two FL algorithms: 1) a cross-center FL algorithm using data distributed across the OPHDIAT centers and 2) a cross-grader FL algorithm using data distributed across the OPHDIAT graders. We explore and assess different FL strategies and compare them to a conventional learning algorithm, namely centralized learning (CL), where all the data is stored in a centralized repository. For the task of referable DR detection, our simulated FL algorithms achieved similar performance to CL, in terms of area under the ROC curve (AUC): AUC =0.9482 for CL, AUC = 0.9317 for cross-center FL and AUC = 0.9522 for cross-grader FL. Our work indicates that the FL algorithm is a viable and reliable framework that can be applied in a screening network.Clinical relevance- Given that data sharing is regarded as an essential component of modern medical research, achieving collaborative learning while protecting sensitive data is key.

Management of diabetic macular oedema in France from 2012 to 2018: The nationwide LANDSCAPE study.

OBJECTIVE: To describe the management of diabetic macular oedema (DME) patients from the entire French population between 2012 and 2018. METHODS: In this retrospective longitudinal study, we identified adults treated for DME from the French population using the exhaustive French National Health Information database (SNDS), and an algorithm based on diagnosis and procedure codes, and reimbursed treatments. RESULTS: Between 2012 and 2018, we identified 53 584 treated DME patients, who were followed for up to 7 years from DME treatment initiation. Optical coherence tomography (OCT) became the predominant imaging tool to diagnose DME. Only 14% of patients consulted a diabetologist or endocrinologist in the 3 months prior to initiating DME treatment, whereas 84% consulted a general practitioner. The percentage of patients consulting an ophthalmologist declined over time, from 97% of patients in Year 1 (median of 9 consultations), to 46% in Year 7 (median of 7 consultations). The median DME treatment duration with an anti-VEGF and/or dexamethasone implant treatment was 9 months; 54% of patients had a treatment duration less than 1 year. First-line treatment was more common with ranibizumab (55% of patients) than with aflibercept (30%), or dexamethasone implant (15%). About 25% of patients who initiated anti-VEGF treatment switched treatment at least once, while 30% of patients who initiated dexamethasone implant switched to anti-VEGF treatment at least once. CONCLUSIONS: French DME patients seem well-monitored by their ophthalmologist, but median DME treatment duration was just 9 months. These results emphasise the challenge to manage and treat patients with DME over the long term.

Towards population-independent, multi-disease detection in fundus photographs.

Independent validation studies of automatic diabetic retinopathy screening systems have recently shown a drop of screening performance on external data. Beyond diabetic retinopathy, this study investigates the generalizability of deep learning (DL) algorithms for screening various ocular anomalies in fundus photographs, across heterogeneous populations and imaging protocols. The following datasets are considered: OPHDIAT (France, diabetic population), OphtaMaine (France, general population), RIADD (India, general population) and ODIR (China, general population). Two multi-disease DL algorithms were developed: a Single-Dataset (SD) network, trained on the largest dataset (OPHDIAT), and a Multiple-Dataset (MD) network, trained on multiple datasets simultaneously. To assess their generalizability, both algorithms were evaluated whenever training and test data originate from overlapping datasets or from disjoint datasets. The SD network achieved a mean per-disease area under the receiver operating characteristic curve (mAUC) of 0.9571 on OPHDIAT. However, it generalized poorly to the other three datasets (mAUC < 0.9). When all four datasets were involved in training, the MD network significantly outperformed the SD network (p = 0.0058), indicating improved generality. However, in leave-one-dataset-out experiments, performance of the MD network was significantly lower on populations unseen during training than on populations involved in training (p < 0.0001), indicating imperfect generalizability.

Prospective, Observational, Multicenter, Real-World Study of the Efficacy, Safety, and Pattern of Use of the Dexamethasone Intravitreal Implant in Diabetic Macular Edema in France: Short-Term Outcomes of LOUVRE 3.

INTRODUCTION: To evaluate real-world efficacy, safety, and treatment patterns with the dexamethasone intravitreal implant (DEX) in diabetic macular edema (DME) in France. METHODS: In this prospective, multicenter, observational, noncomparative, post-reimbursement study, consecutively enrolled patients with DME had a baseline evaluation on day 0. Those treated with DEX on day 0 were to be reevaluated at week 6 and months 6, 12, 18, and 24. DEX retreatment and/or alternative therapies were allowed during follow-up. The primary outcome measure was the maximum best corrected visual acuity (BCVA) gain from baseline during follow-up. Secondary outcome measures included time to maximum BCVA gain, patients (%) with prespecified BCVA gains from baseline at each visit, maximum central retinal thickness (CRT) reduction from baseline, patients (%) with CRT reduction ≥ 20% from baseline at each visit, patients (%) with DME resolution (per investigator judgement), and adverse events (AEs). RESULTS: Of 112 patients/eyes with DME for 3.5 years (mean) at baseline, 80 (including 86.1% previously treated) received DEX on day 0 and were analyzed for efficacy. Early study termination precluded collection of ≥ 12-month efficacy data. Patients received 1.4 DEX injections over 8.3 months (averages). The maximum BCVA gain from baseline was 3.6 letters, reached after 77.2 days (averages); 24.6% (week 6) and 15.0% (month 6) of patients experienced ≥ 10-letter BCVA gains from baseline. The mean maximum CRT reduction from baseline was -146.4 µm; 61.4% (week 6) and 36.0% (month 6) of patients had CRT reductions ≥ 20% from baseline, and 68.1% reported DME resolution at least once during follow-up. Ocular hypertension (n = 8, 12.1%) was the most frequent treatment-related AE. CONCLUSIONS: LOUVRE 3 confirmed that DEX improves BCVA and CRT, even in a patient population that had predominantly received DEX before enrollment in the study, and showed that DME resolution was observed during follow-up. DEX tolerability was consistent with published data, supporting treatment benefits in DME. GOV IDENTIFIER: NCT03003416.

Real-World Outcomes of Ranibizumab Treatment in French Patients with Visual Impairment due to Macular Edema Secondary to Retinal Vein Occlusion: 24-Month Results from the BOREAL-RVO Study.

INTRODUCTION: Information about real-world ranibizumab use is needed to optimize treatment of macular edema secondary to retinal vein occlusion (RVO). The BOREAL-RVO study assessed treatment use, effectiveness, and safety of 24-month treatment with ranibizumab 0.5 mg in patients with visual impairment due to macular edema secondary to RVO in a real-world setting. METHODS: This was a multicenter, post-authorization, observational study in France, including patients starting ranibizumab for RVO. Primary endpoint was mean change from baseline in best-corrected visual acuity (BCVA) at month 6. Secondary endpoints were mean changes from baseline in BCVA at month 24 and central retinal thickness (CRT) at months 6 and 24, and treatment use in real-world setting. RESULTS: 226 branch RVO (BRVO) and 196 central RVO (CRVO) patients were enrolled; 71.7% and 70.9% completed the 24-month follow-up, respectively. In BRVO, mean (SD) baseline BCVA was 55.2 (18.7) letters, with gains of 14.3 (13.7), 14.1 (16.5), 13.0 (17.5), and 11.4 (20.1) letters at months 3, 6, 12, and 24, respectively. In CRVO, mean (SD) baseline BCVA was 40.4 (25.6) letters, with gains of 16.0 (21.2), 9.5 (25.4), 9.2 (27.7), and 8.3 (23.8) letters at months 3, 6, 12, and 24, respectively. At month 24, 52% of BRVO and 41% of CRVO patients had gains of 15 or more letters. In BRVO, mean (SD) CRT values at baseline and months 3, 6, 12, and 24 were 550 (175), 315 (104), 343 (122), 335 (137), and 340 (105) µm. In CRVO, mean (SD) CRT values at baseline and months 3, 6, 12, and 24 were 643 (217), 327 (152), 400 (203), 379 (175), and 348 (161) µm. On average, BRVO patients had 3.8 injections for 6.9 visits by month 6, and 7.2 injections for 19.7 visits by month 24. CRVO patients had 2.7 injections for 4.2 visits by month 6 and 7.1 injections for 21.1 visits by month 24. Factors predictive of better BCVA gain at month 6 were age under 60 at baseline, lower baseline BCVA and BCVA gain at month 3. There were no new safety findings. CONCLUSION: Major improvements in BCVA and CRT were observed at month 3 after the induction phase and then were sustained up to month 24, with a slight decrease, probably due to under-treatment. This study demonstrated ranibizumab to be a safe and effective treatment for BRVO and CRVO in the real-world setting, although more regular or proactive treatment could further improve outcomes.

Evaluation of three hemagglutinin-based vaccines for the experimental control of a panzootic clade 2.3.4.4b A(H5N8) high pathogenicity avian influenza virus in mule ducks.

In France during winter 2016-2017, 487 outbreaks of clade 2.3.4.4b H5N8 subtype high pathogenicity (HP) avian influenza A virus (AIV) infections were detected in poultry and captive birds. During this epizootic, HPAIV A/decoy duck/France/161105a/2016 (H5N8) was isolated and characterized in an experimental infection transmission model in conventional mule ducks. To investigate options to possibly protect such ducks against this HPAIV, three vaccines were evaluated in controlled conditions. The first experimental vaccine was derived from the hemagglutinin gene of another clade 2.3.4.4b A(H5N8) HPAIV. It was injected at three weeks of age, either alone (Vac1) or after a primer injection at day-old (Vac1 + boost). The second vaccine (Vac2) was a commercial bivalent adjuvanted vaccine containing an expressed hemagglutinin modified from a clade 2.3.2 A(H5N1) HPAIV. Vac2 was administered as a single injection at two weeks of age. The third experimental vaccine (Vac3) also incorporated a homologous 2.3.4.4b H5 HA gene and was administered as a single injection at three weeks of age. Ducks were challenged with HPAIV A/decoy duck/France/161105a/2016 (H5N8) at six weeks of age. Post-challenge virus excretion was monitored in vaccinated and control birds every 2-3 days for two weeks using real-time reverse-transcription polymerase chain reaction and serological analyses (haemagglutination inhibition test against H5N8, H5 ELISA and AIV ELISA) were performed. Vac1 abolished oropharyngeal and cloacal shedding to almost undetectable levels, whereas Vac3 abolished cloacal shedding only (while partially reducing respiratory shedding) and Vac2 only partly reduced the respiratory and intestinal excretion of the challenge virus. These results provided relevant insights in the immunogenicity of recombinant H5 vaccines in mule ducks, a rarely investigated hybrid between Pekin and Muscovy duck species that has played a critical role in the recent H5 HPAI epizootics in France.

IMPACT study: Impact of adherence to anti-VEGF intravitreal injections for macular disease during COVID 19-related confinement in France.

PURPOSE: The aim of this study was to evaluate the impact of adherence to French coronavirus disease 2019 (COVID 19)-related guidelines for intravitreal injection (IVI) practice on the visual outcomes of patients treated with anti-vascular endothelial growth factor (VEGF) agents for macular diseases during the first lockdown period. METHODS: Observational multicentre study including all patients from 18 centres with an IVI initially planned during the lockdown. Visual acuity (VA, ETDRS) was recorded at 1 and 4 months after lockdown. French COVID 19-related guidelines recommended maintaining IVI practice. We defined three groups of patients: A, adherent to guidelines; NA+, non-adherent with delayed IVIs; and NA-, non-adherent without IVIs performed during the lockdown. Risk factors for non-adherence and visual loss were studied. RESULTS: A total of 3020 eyes of 3020 patients, aged 77.8 ± 11.6 years, 59.8% women, were included. 59.3% were non-adherent(46.7% NA+, 12.6% NA-). A smaller decrease in VA at 4 months was observed in the A group than the NA+ and NA- group (-0.2 ± 6.7, -0.3 ± 6.9 and -1.5 ± 6.9, respectively [p < 0.001]). Factors associated with non-adherence were in multivariable analysis, older age, hospital practice, low-density population areas, high viral incidence areas, longer intervals between injection and treat and extent protocol. Factors associated with visual loss at 4 months in multivariable analysis were, being in the NA- group, older age, T&E and fixed regimens. CONCLUSION: Strict adherence to guidelines was associated with better visual outcome, although most of our patients did not attend as planned. Identification of patients at risk could help in the future in case of a new pandemic lockdown.

Epidemiology of Treated Diabetes Ocular Complications in France 2008-2018-The LANDSCAPE French Nationwide Study.

AIM: LANDSCAPE aimed to estimate the annual incidence and prevalence of treated diabetic macular edema (DME) and proliferative diabetic retinopathy (PDR) between 2008 and 2018. METHODS: This French nationwide observational study used data from the French National Health Insurance Databases covering 99% of the French population. Data about healthcare consumption were used to identify adults treated with anti-VEGFs or dexamethasone implants (for DME) and with pan-retinal photocoagulation (for PDR). All French patients newly treated between 2008 and 2018 were included. Incidence and prevalence of treated DME and PDR were estimated for the age-matched general population and the population with diabetes in France. Sociodemographic characteristics and medical history were described in both populations. RESULTS: We identified 53,584 treated DME patients and 127,273 treated PDR patients between 2008 and 2018, and 11,901 DME and 11,996 PDR new incident patients in 2018. The treated DME incidence in 2018 was 2.5 per 10,000 in the general population and 37.3 per 10,000 in the population with diabetes. Prevalence in 2018 was 9.5 and 143.7 per 10,000 in the respective populations. Treated PDR incidence in 2018 was 2.3 per 10,000 in the general population and 31.2 per 10,000 in the population with diabetes. Prevalence in 2018 was 19.9 and 270.3 per 10,000 in the respective populations. Incidence and prevalence were not age-dependent. Incidence of treated PDR incidence was relatively stable from 2008-2018. Incidence of treated DME incidence rose from 2012-2018, probably due to widening access to newly available treatments, such as anti-VEGFs. CONCLUSIONS: We provide exhaustive nationwide data on the incidence and prevalence of treated diabetic ocular complications in France over a 10-year period.

Management of Patients with Diabetic Macular Edema Switched from Dexamethasone Intravitreal Implant to Fluocinolone Acetonide Intravitreal Implant.

To assess anatomical and functional outcomes after switching from dexamethasone implant (DEXi) to fluocinolone acetonide implant (FAci) in 113 diabetic macular edema eyes, a multicentric retrospective observational study was conducted. Seventy-five eyes (73.5%) were switched 1−8 weeks after their last DEXi. The mean best-corrected visual acuity improved to 59.8 letters at month 4 and remained stable during the follow-up. The mean central macular thickness (CMT) significantly decreased during the follow-up, with a minimum of 320.9 µm at month 3. The baseline CMT was higher in eyes that received the last DEXi >8 weeks versus <8 weeks before the first FAci (p < 0.021). After FAci injection, additional treatments were needed in 37 (32.7%) eyes. A longer diabetes duration (p = 0.009), a longer time between the last DEXi and the first FAci (p = 0.035), and a high baseline CMT (p = 0.003) were risk factors for additional treatments. The mean intraocular pressure was <19 mmHg at all timepoints, with no difference between eyes receiving the last DEXi ≤8 weeks or >8 weeks before the switch. Switching from DEXi to FAci in DME is effective and safe. A short time between the last DEXi and the first FAci reduced CMT fluctuations and the need for early additional treatments.

Two-year outcomes of the APOLLON observational study of intravitreal aflibercept monotherapy in France in patients with diabetic macular edema.

APOLLON (NCT02924311) was a prospective observational study to evaluate the effectiveness of intravitreal aflibercept (IVT-AFL) treatment of diabetic macular edema (DME) over 24 months in routine clinical practice in France. The primary endpoint was mean change from baseline in best-corrected visual acuity (BCVA; Early Treatment Diabetic Retinopathy Study letters) by 12 months, and safety was monitored throughout the study. Of 402 patients enrolled across 61 participating clinics and hospitals in France, 168 patients were followed for at least 24 months and included in the effectiveness analyses (79 treatment-naïve and 89 previously treated). After 24 months of IVT-AFL treatment, the mean (± standard deviation [SD]) change in BCVA from baseline was + 6.5 (± 10.7) letters in treatment-naïve patients (p < 0.001) and + 1.6 (± 17.0) letters in previously treated patients (p = 0.415) from a baseline of 63.8 (± 13.6) and 60.5 (± 16.5) letters. The mean number of IVT-AFL treatments over 24 months was 11.3 (± 4.9) and 11.9 (± 4.7) for treatment-naïve and previously treated patients. This final analysis of the APOLLON study indicated that following 24 months of IVT-AFL treatment in routine clinical practice in France, treatment-naïve patients with DME achieved significant gains in visual acuity and previously treated patients maintained prior visual acuity gains.Trial registration number: NCT02924311.

Multiple independent introductions of highly pathogenic avian influenza H5 viruses during the 2020-2021 epizootic in France.

During winter 2020-2021, France and other European countries were severely affected by highly pathogenic avian influenza H5 viruses of the Gs/GD/96 lineage, clade 2.3.4.4b. In total, 519 cases occurred, mainly in domestic waterfowl farms in Southwestern France. Analysis of viral genomic sequences indicated that 3 subtypes of HPAI H5 viruses were detected (H5N1, H5N3, H5N8), but most French viruses belonged to the H5N8 subtype genotype A, as Europe. Phylogenetic analyses of HPAI H5N8 viruses revealed that the French sequences were distributed in 9 genogroups, suggesting 9 independent introductions of H5N8 from wild birds, in addition to the 2 introductions of H5N1 and H5N3.

Concomitant NA and NS deletion on avian Influenza H3N1 virus associated with hen mortality in France in 2019.

An H3N1 avian influenza virus was detected in a laying hens farm in May 2019 which had experienced 25% mortality in Northern France. The complete sequencing of this virus showed that all segment sequences belonged to the Eurasian lineage and were phylogenetically very close to many of the Belgian H3N1 viruses detected in 2019. The French virus presented two genetic particularities with NA and NS deletions that could be related to virus adaptation from wild to domestic birds and could increase virulence, respectively. Molecular data of H3N1 viruses suggest that these two deletions occurred at two different times.

Intraocular Pressure Changes After Intravitreal Fluocinolone Acetonide Implant: Results from Four European Countries.

INTRODUCTION: The 0.19 mg fluocinolone acetonide (FAc) intravitreal implant delivers a continuous intravitreal corticosteroid dose for the treatment of refractory diabetic macular oedema (DMO). The aim of this study was to assess the impact of an FAc intravitreal implant on intraocular pressure (IOP). METHODS: We retrospectively collected anonymised data on the patients' characteristics, DMO treatment, and IOP and IOP-lowering treatments before and after the FAc intravitreal implant between September 2013 and March 2020 in several European centres. RESULTS: A total of 221 eyes from 179 patients were included. The mean follow-up duration was 13.4 (± 12.5, range 2.4-33.5) months. Overall, 194 eyes (88.2%) had received an intravitreal dexamethasone injection before the FAc intravitreal implant. For 25 eyes (11.3%) there was a history of glaucoma, and 52 eyes (23.5%) had previous IOP-lowering treatment. Mean IOP before injection was 14.7 (3.4) mmHg and increased to 16.9 (3.7) mmHg 12 months after injection (P < 0.0001). During follow-up, 55 eyes (24.9%) required the addition or initiation of topical IOP-lowering medication, only one patient (0.5%) had laser trabeculoplasty and one patient (0.5%) a minimally invasive glaucoma surgery, and no patient required incisional IOP-lowering surgery. CONCLUSION: The FAc intravitreal implant led to substantial IOP elevation. This elevation was monitored most of the time with addition or initiation of topical IOP-lowering medication.

Avian influenza outbreaks: evaluating the efficacy of cleaning and disinfection of vehicles and transport crates.

In 2021, France faced large avian influenza outbreaks, like in 2016 and 2017. Controlling these outbreaks required the preventive depopulation of a large number of duck farms. A previous study in 2017 showed that the quality of decontamination of trucks and transport crates used for depopulation was often insufficient. A new study was then set up to evaluate cleaning and disinfection (C&D) of trucks and crates used for duck depopulation and whether practices had changed since 2017. Three methods were used to assess decontamination: 1) detection of avian influenza virus (AIV) genome, 2) visual inspection of cleanliness, and 3) microbial counts, considering that 2 and 3 are commonly used in abattoirs. Another objective of the study was to evaluate the correlation between results obtained with the 3 methods. In 5 abattoirs, 8 trucks and their crates were sampled by swabbing to detect AIV genome by rRT-PCR before and after decontamination. Visual cleanliness scores and coliform counts were also determined on crates after C&D. Trucks and crates were decontaminated according to the abattoirs' protocols. Before C&D, 3 quarters of crates (59/79) and 7 of 8 trucks were positive for AIV genome. C&D procedures were reinforced in 2021 compared to 2017; use of detergent solution and warm water were more common. Nevertheless, 28% of the crates were positive for AIV genome after C&D, despite the fact that cleaning scores and microbiological counts were satisfactory for 84% and 91% of the crates, respectively. No correlation was observed between results for AIV genome detection and results from visual control or from coliform counts. Abattoirs are encouraged to use environmental sampling coupled with AIV genome detection to monitor the quality of cleaning and disinfection of trucks and crates during AI outbreaks. Reinforcement of biosecurity measures at abattoirs is still needed to avoid residual contamination of the equipment and cross-contamination during the decontamination process.

Automatic Screening for Ocular Anomalies Using Fundus Photographs.

SIGNIFICANCE: Screening for ocular anomalies using fundus photography is key to prevent vision impairment and blindness. With the growing and aging population, automated algorithms that can triage fundus photographs and provide instant referral decisions are relevant to scale-up screening and face the shortage of ophthalmic expertise. PURPOSE: This study aimed to develop a deep learning algorithm that detects any ocular anomaly in fundus photographs and to evaluate this algorithm for "normal versus anomalous" eye examination classification in the diabetic and general populations. METHODS: The deep learning algorithm was developed and evaluated in two populations: the diabetic and general populations. Our patient cohorts consist of 37,129 diabetic patients from the OPHDIAT diabetic retinopathy screening network in Paris, France, and 7356 general patients from the OphtaMaine private screening network, in Le Mans, France. Each data set was divided into a development subset and a test subset of more than 4000 examinations each. For ophthalmologist/algorithm comparison, a subset of 2014 examinations from the OphtaMaine test subset was labeled by a second ophthalmologist. First, the algorithm was trained on the OPHDIAT development subset. Then, it was fine-tuned on the OphtaMaine development subset. RESULTS: On the OPHDIAT test subset, the area under the receiver operating characteristic curve for normal versus anomalous classification was 0.9592. On the OphtaMaine test subset, the area under the receiver operating characteristic curve was 0.8347 before fine-tuning and 0.9108 after fine-tuning. On the ophthalmologist/algorithm comparison subset, the second ophthalmologist achieved a specificity of 0.8648 and a sensitivity of 0.6682. For the same specificity, the fine-tuned algorithm achieved a sensitivity of 0.8248. CONCLUSIONS: The proposed algorithm compares favorably with human performance for normal versus anomalous eye examination classification using fundus photography. Artificial intelligence, which previously targeted a few retinal pathologies, can be used to screen for ocular anomalies comprehensively.

Prevalence, severity stages, and risk factors of diabetic retinopathy in 1464 adult patients with type 1 diabetes.

PURPOSE: To determine the prevalence of diabetic retinopathy (DR) and its risk factors in adult type 1 diabetes (T1D) patients METHODS: In this cross-sectional study, all T1D patients followed in the University Center for Diabetes and its Complications of Lariboisière Hospital (Paris, France) between January 2017 and February 2019 were included. Ophthalmologic and systemic data were collected from electronic records. The association between DR (and each grade) and associated factors were estimated by univariate and multivariate analyses using logistic regression models. RESULTS: A total of 1464 patients (46.2% of women, mean age: 42.2 ± 15.8 years) were included. The mean hemoglobin A1c (HbA1c) was 7.8 ± 1.7% and the mean diabetes duration was 20.5 ± 13.5 years. DR prevalence was 50.1% (47.4-52.6) and the prevalence of mild, moderate, and severe non-proliferative DR and proliferative DR was 19.1%, 9.4%, 3.9%, and 17.6%, respectively. DR was significantly associated with male gender, an older age, former and current smoking status, a higher BMI, the presence of nephropathy and neuropathy, higher HBA1c, and longer diabetes duration. Patients with HbA1c > 10% had an adjusted odds ratio (OR) of 3.25 (1.77-6.01) of having DR compared to patients with HbA1c < 6.5%. Patients with a diabetes duration > 30 years had an adjusted OR of 24.87 (14.82-42.67) higher of having DR compared to patients with a diabetes duration < 10 years. CONCLUSION: In this study, 50.1% of adult T1D patients had DR and 17.6% had proliferative DR. Diabetes duration and HbA1c were major risk factors.

ExplAIn: Explanatory artificial intelligence for diabetic retinopathy diagnosis.

In recent years, Artificial Intelligence (AI) has proven its relevance for medical decision support. However, the "black-box" nature of successful AI algorithms still holds back their wide-spread deployment. In this paper, we describe an eXplanatory Artificial Intelligence (XAI) that reaches the same level of performance as black-box AI, for the task of classifying Diabetic Retinopathy (DR) severity using Color Fundus Photography (CFP). This algorithm, called ExplAIn, learns to segment and categorize lesions in images; the final image-level classification directly derives from these multivariate lesion segmentations. The novelty of this explanatory framework is that it is trained from end to end, with image supervision only, just like black-box AI algorithms: the concepts of lesions and lesion categories emerge by themselves. For improved lesion localization, foreground/background separation is trained through self-supervision, in such a way that occluding foreground pixels transforms the input image into a healthy-looking image. The advantage of such an architecture is that automatic diagnoses can be explained simply by an image and/or a few sentences. ExplAIn is evaluated at the image level and at the pixel level on various CFP image datasets. We expect this new framework, which jointly offers high classification performance and explainability, to facilitate AI deployment.