A deep-learning system predicts glaucoma incidence and progression using retinal photographs.
The Journal of clinical investigation
BackgroundDeep learning has been widely used for glaucoma diagnosis. However, there is no clinically validated algorithm for glaucoma incidence and progression prediction. This study aims to develop a clinically feasible deep-learning system for predicting and stratifying the risk of glaucoma onset and progression based on color fundus photographs (CFPs), with clinical validation of performance in external population cohorts.MethodsWe established data sets of CFPs and visual fields collected from longitudinal cohorts. The mean follow-up duration was 3 to 5 years across the data sets. Artificial intelligence (AI) models were developed to predict future glaucoma incidence and progression based on the CFPs of 17,497 eyes in 9346 patients. The area under the receiver operating characteristic (AUROC) curve, sensitivity, and specificity of the AI models were calculated with reference to the labels provided by experienced ophthalmologists. Incidence and progression of glaucoma were determined based on longitudinal CFP images or visual fields, respectively.ResultsThe AI model to predict glaucoma incidence achieved an AUROC of 0.90 (0.81-0.99) in the validation set and demonstrated good generalizability, with AUROCs of 0.89 (0.83-0.95) and 0.88 (0.79-0.97) in external test sets 1 and 2, respectively. The AI model to predict glaucoma progression achieved an AUROC of 0.91 (0.88-0.94) in the validation set, and also demonstrated outstanding predictive performance with AUROCs of 0.87 (0.81-0.92) and 0.88 (0.83-0.94) in external test sets 1 and 2, respectively.ConclusionOur study demonstrates the feasibility of deep-learning algorithms in the early detection and prediction of glaucoma progression.FUNDINGNational Natural Science Foundation of China (NSFC); the High-level Hospital Construction Project, Zhongshan Ophthalmic Center, Sun Yat-sen University; the Science and Technology Program of Guangzhou, China (2021), the Science and Technology Development Fund (FDCT) of Macau, and FDCT-NSFC.
10.1172/JCI157968
Deep-learning models for the detection and incidence prediction of chronic kidney disease and type 2 diabetes from retinal fundus images.
Zhang Kang,Liu Xiaohong,Xu Jie,Yuan Jin,Cai Wenjia,Chen Ting,Wang Kai,Gao Yuanxu,Nie Sheng,Xu Xiaodong,Qin Xiaoqi,Su Yuandong,Xu Wenqin,Olvera Andrea,Xue Kanmin,Li Zhihuan,Zhang Meixia,Zeng Xiaoxi,Zhang Charlotte L,Li Oulan,Zhang Edward E,Zhu Jie,Xu Yiming,Kermany Daniel,Zhou Kaixin,Pan Ying,Li Shaoyun,Lai Iat Fan,Chi Ying,Wang Changuang,Pei Michelle,Zang Guangxi,Zhang Qi,Lau Johnson,Lam Dennis,Zou Xiaoguang,Wumaier Aizezi,Wang Jianquan,Shen Yin,Hou Fan Fan,Zhang Ping,Xu Tao,Zhou Yong,Wang Guangyu
Nature biomedical engineering
Regular screening for the early detection of common chronic diseases might benefit from the use of deep-learning approaches, particularly in resource-poor or remote settings. Here we show that deep-learning models can be used to identify chronic kidney disease and type 2 diabetes solely from fundus images or in combination with clinical metadata (age, sex, height, weight, body-mass index and blood pressure) with areas under the receiver operating characteristic curve of 0.85-0.93. The models were trained and validated with a total of 115,344 retinal fundus photographs from 57,672 patients and can also be used to predict estimated glomerulal filtration rates and blood-glucose levels, with mean absolute errors of 11.1-13.4 ml min per 1.73 m and 0.65-1.1 mmol l, and to stratify patients according to disease-progression risk. We evaluated the generalizability of the models for the identification of chronic kidney disease and type 2 diabetes with population-based external validation cohorts and via a prospective study with fundus images captured with smartphones, and assessed the feasibility of predicting disease progression in a longitudinal cohort.
10.1038/s41551-021-00745-6
Artificial intelligence for pediatric ophthalmology.
Reid Julia E,Eaton Eric
Current opinion in ophthalmology
PURPOSE OF REVIEW:Despite the impressive results of recent artificial intelligence applications to general ophthalmology, comparatively less progress has been made toward solving problems in pediatric ophthalmology using similar techniques. This article discusses the unique needs of pediatric patients and how artificial intelligence techniques can address these challenges, surveys recent applications to pediatric ophthalmology, and discusses future directions. RECENT FINDINGS:The most significant advances involve the automated detection of retinopathy of prematurity, yielding results that rival experts. Machine learning has also been applied to the classification of pediatric cataracts, prediction of postoperative complications following cataract surgery, detection of strabismus and refractive error, prediction of future high myopia, and diagnosis of reading disability. In addition, machine learning techniques have been used for the study of visual development, vessel segmentation in pediatric fundus images, and ophthalmic image synthesis. SUMMARY:Artificial intelligence applications could significantly benefit clinical care by optimizing disease detection and grading, broadening access to care, furthering scientific discovery, and improving clinical efficiency. These methods need to match or surpass physician performance in clinical trials before deployment with patients. Owing to the widespread use of closed-access data sets and software implementations, it is difficult to directly compare the performance of these approaches, and reproducibility is poor. Open-access data sets and software could alleviate these issues and encourage further applications to pediatric ophthalmology.
10.1097/ICU.0000000000000593
Artificial Intelligence for Screening of Multiple Retinal and Optic Nerve Diseases.
JAMA network open
Importance:The lack of experienced ophthalmologists limits the early diagnosis of retinal diseases. Artificial intelligence can be an efficient real-time way for screening retinal diseases. Objective:To develop and prospectively validate a deep learning (DL) algorithm that, based on ocular fundus images, recognizes numerous retinal diseases simultaneously in clinical practice. Design, Setting, and Participants:This multicenter, diagnostic study at 65 public medical screening centers and hospitals in 19 Chinese provinces included individuals attending annual routine medical examinations and participants of population-based and community-based studies. Exposures:Based on 120 002 ocular fundus photographs, the Retinal Artificial Intelligence Diagnosis System (RAIDS) was developed to identify 10 retinal diseases. RAIDS was validated in a prospective collected data set, and the performance between RAIDS and ophthalmologists was compared in the data sets of the population-based Beijing Eye Study and the community-based Kailuan Eye Study. Main Outcomes and Measures:The performance of each classifier included sensitivity, specificity, accuracy, F1 score, and Cohen κ score. Results:In the prospective validation data set of 208 758 images collected from 110 784 individuals (median [range] age, 42 [8-87] years; 115 443 [55.3%] female), RAIDS achieved a sensitivity of 89.8% (95% CI, 89.5%-90.1%) to detect any of 10 retinal diseases. RAIDS differentiated 10 retinal diseases with accuracies ranging from 95.3% to 99.9%, without marked differences between medical screening centers and geographical regions in China. Compared with retinal specialists, RAIDS achieved a higher sensitivity for detection of any retinal abnormality (RAIDS, 91.7% [95% CI, 90.6%-92.8%]; certified ophthalmologists, 83.7% [95% CI, 82.1%-85.1%]; junior retinal specialists, 86.4% [95% CI, 84.9%-87.7%]; and senior retinal specialists, 88.5% [95% CI, 87.1%-89.8%]). RAIDS reached a superior or similar diagnostic sensitivity compared with senior retinal specialists in the detection of 7 of 10 retinal diseases (ie, referral diabetic retinopathy, referral possible glaucoma, macular hole, epiretinal macular membrane, hypertensive retinopathy, myelinated fibers, and retinitis pigmentosa). It achieved a performance comparable with the performance by certified ophthalmologists in 2 diseases (ie, age-related macular degeneration and retinal vein occlusion). Compared with ophthalmologists, RAIDS needed 96% to 97% less time for the image assessment. Conclusions and Relevance:In this diagnostic study, the DL system was associated with accurately distinguishing 10 retinal diseases in real time. This technology may help overcome the lack of experienced ophthalmologists in underdeveloped areas.
10.1001/jamanetworkopen.2022.9960
Risk prediction of dysthyroid optic neuropathy based on CT imaging features combined the bony orbit with the soft tissue structures.
Frontiers in medicine
Purpose:To analyze computed tomographic (CT) imaging features of patients with dysthyroid optic neuropathy (DON) retrospectively and deduce a more appropriate predictive model. Methods:The CT scans and medical records of 60 patients with clinically proven Graves' ophthalmopathy (GO) with (26 women and 10 men) and without DON (16 women and 8 men) were retrospectively reviewed, and 20 age- and sex-matched control participants (12 women and 8 men) were enrolled consecutively. The bony orbit [orbital rim angle (ORA), medial and lateral orbital wall angles (MWA and LWA), orbital apex angle (OAA), and length of the lateral orbital wall (LWL)], and the soft tissue structures [maximum extraocular muscle diameters (Max EOMD), muscle diameter index (MDI), medial and lateral rectus bulk from inter-zygomatic line (MRIZL and LRIZL), proptosis, intraorbital optic nerve stretching length (IONSL), superior ophthalmic vein diameter (SOVD), apical crowding, and presence of intracranial fat prolapse] were assessed on a clinical workstation. The CT features among groups were compared, and a multivariate logistic regression analysis was performed to evaluate the predictive features of DON. Results:All bony orbital angle indicators, except ORA ( = 0.461), were statistically different among the three groups (all < 0.05). The values of MWA, LWA, OAA, and LWL were larger in the orbits with the DON group than in the orbits without the DON group (all < 0.05). The MDI, MRIZL, proptosis, IONSL, and SOVD were statistically significantly different among the three groups (all < 0.05), in which the orbits with the DON group were significantly higher than the orbits without the DON group and control group. The apical crowding was more severe in the orbits with the DON group than in the orbits without the DON group ( = 0.000). There were no significant differences in the LRIZL and the presence of intracranial fat prolapse (all > 0.05). The multivariate regression analysis showed that the MWA, MDI, and SOVD were the independent factors predictive of DON. The sensitivity and specificity for the presence of DON by combining these three indicators were 89 and 83%, respectively. Conclusion:Bone and soft tissue CT features are useful in the risk prediction of DON, especially the MWA, MDI, and SOVD were the independent factors predictive of DON.
10.3389/fmed.2022.936819
The risk factors of neuropathic pain in neuromyelitis optica spectrum disorder: a retrospective case-cohort study.
BMC neurology
BACKGROUND:Neuropathic pain is a common complication in neuromyelitis optica spectrum disorder (NMOSD), which seriously affects the quality of life of NMOSD patients, with no satisfactory treatment. And risk factors of neuropathic pain are still uncertain. OBJECTIVE:To investigate the risk factors of neuropathic pain in a NMOSD cohort. MATERIALS AND METHODS:Our study was a retrospective case-cohort study, the patients diagnosed with NMOSD in the Department of Neurology from the Second Affiliated Hospital of Guangzhou University of Chinese Medicine from January 2011 to October 2021 were screened. Inclusion criteria were: (1) patients diagnosed as NMOSD according to the International Panel for NMO Diagnosis (IPND) criteria, (2) the aquaporin-4 immunoglobulin G antibodies (AQP4-IgG) test was performed. Patients without AQP4-IgG antibody were excluded. Clinical data, including sex, age of the first onset, symptoms of the first episode including neuropathic pain and attack types, localization of lesions of the first episode on Magnetic Resonance Imaging (MRI), Extended disability status Scale (EDSS) of the first onset, treatment of immunosuppression in the first acute phase, disease modifying therapy (DMT), treatment of neuropathic pain and APQ4-IgG status were collected from the hospital system database. Neuropathic pain was defined according to the International Association for the Study of Pain criteria and was described as "pain arising as a direct consequence of a lesion or disease affecting the somatosensory system". RESULTS:One hundred nineteen patients were screened and finally 86 patients fulfilling the inclusion and exclusion criteria were enrolled in our study. The prevalence of neuropathic pain in patients with NMOSD was 43.0%. Univariate analysis showed that the factors associated with neuropathic pain were the age at the onset, the attack type of optic neuritis, the attack type of myelitis, length of spinal cord involvement, localization of thoracic lesion, optic lesion, upper thoracic lesions, lower thoracic lesions, extended spinal cord lesions (≥ 3 spinal lesions), extended thoracic lesions (≥ 4 thoracic lesions), intravenous immunoglobulin and mycophenolate mofetil. Multivariate regression analysis showed that extended thoracic lesions (OR 20.21 [1.18-346.05], P = 0.038) and age (OR 1.35 (1-1.81) P = 0.050) were independently associated with neuropathic pain among NMOSD patients and that gender (OR 12.11 (0.97-151.64) P = 0.053) might be associated with neuropathic pain among NMOSD patients. CONCLUSION:Extended thoracic lesions (≥ 4 thoracic lesions), age and gender might be independent risk factors of neuropathic pain among patients with NMOSD. However, with a small sample size and predominantly female, caution must be applied and these results need validating in further cohorts.
10.1186/s12883-022-02841-9
A multivariate analysis and statistical model for predicting visual acuity and keratometry one year after cross-linking for keratoconus.
Wisse Robert P L,Godefrooij Daniël A,Soeters Nienke,Imhof Saskia M,Van der Lelij Allegonda
American journal of ophthalmology
PURPOSE:To investigate putative prognostic factors for predicting visual acuity and keratometry 1 year following corneal cross-linking (CXL) for treating keratoconus. DESIGN:Prospective cohort study. METHODS:This study included all consecutively treated keratoconus patients (102 eyes) in 1 academic treatment center, with minimal 1-year follow-up following CXL. Primary treatment outcomes were corrected distance visual acuity (logMAR CDVA) and maximum keratometry (K(max)). Univariable analyses were performed to determine correlations between baseline parameters and follow-up measurements. Correlating factors (P ≤ .20) were then entered into a multivariable linear regression analysis, and a model for predicting CDVA and K(max) was created. RESULTS:Atopic constitution, positive family history, and smoking were not independent factors affecting CXL outcomes. Multivariable analysis identified cone eccentricity as a major factor for predicting K(max) outcome (ß coefficient = 0.709, P = .02), whereas age, sex, and baseline keratometry were not independent contributors. Posttreatment visual acuity could be predicted based on pretreatment visual acuity (ß coefficient = -0.621, P < .01, R(2) = 0.45). Specifically, a low visual acuity predicts visual improvement. A prediction model for K(max) did not accurately estimate treatment outcomes (R(2) = 0.15). CONCLUSIONS:Our results confirm the role of cone eccentricity with respect to the improvement of corneal curvature following CXL. Visual acuity outcome can be predicted accurately based on pretreatment visual acuity. Age, sex, and K(max) are debated as independent factors for predicting the outcome of treating keratoconus with CXL.
10.1016/j.ajo.2013.11.001
Development and Validation of a Clinical Risk Score to Predict the Occurrence of Critical Illness in Hospitalized Patients With COVID-19.
Liang Wenhua,Liang Hengrui,Ou Limin,Chen Binfeng,Chen Ailan,Li Caichen,Li Yimin,Guan Weijie,Sang Ling,Lu Jiatao,Xu Yuanda,Chen Guoqiang,Guo Haiyan,Guo Jun,Chen Zisheng,Zhao Yi,Li Shiyue,Zhang Nuofu,Zhong Nanshan,He Jianxing,
JAMA internal medicine
Importance:Early identification of patients with novel coronavirus disease 2019 (COVID-19) who may develop critical illness is of great importance and may aid in delivering proper treatment and optimizing use of resources. Objective:To develop and validate a clinical score at hospital admission for predicting which patients with COVID-19 will develop critical illness based on a nationwide cohort in China. Design, Setting, and Participants:Collaborating with the National Health Commission of China, we established a retrospective cohort of patients with COVID-19 from 575 hospitals in 31 provincial administrative regions as of January 31, 2020. Epidemiological, clinical, laboratory, and imaging variables ascertained at hospital admission were screened using Least Absolute Shrinkage and Selection Operator (LASSO) and logistic regression to construct a predictive risk score (COVID-GRAM). The score provides an estimate of the risk that a hospitalized patient with COVID-19 will develop critical illness. Accuracy of the score was measured by the area under the receiver operating characteristic curve (AUC). Data from 4 additional cohorts in China hospitalized with COVID-19 were used to validate the score. Data were analyzed between February 20, 2020 and March 17, 2020. Main Outcomes and Measures:Among patients with COVID-19 admitted to the hospital, critical illness was defined as the composite measure of admission to the intensive care unit, invasive ventilation, or death. Results:The development cohort included 1590 patients. the mean (SD) age of patients in the cohort was 48.9 (15.7) years; 904 (57.3%) were men. The validation cohort included 710 patients with a mean (SD) age of 48.2 (15.2) years, and 382 (53.8%) were men and 172 (24.2%). From 72 potential predictors, 10 variables were independent predictive factors and were included in the risk score: chest radiographic abnormality (OR, 3.39; 95% CI, 2.14-5.38), age (OR, 1.03; 95% CI, 1.01-1.05), hemoptysis (OR, 4.53; 95% CI, 1.36-15.15), dyspnea (OR, 1.88; 95% CI, 1.18-3.01), unconsciousness (OR, 4.71; 95% CI, 1.39-15.98), number of comorbidities (OR, 1.60; 95% CI, 1.27-2.00), cancer history (OR, 4.07; 95% CI, 1.23-13.43), neutrophil-to-lymphocyte ratio (OR, 1.06; 95% CI, 1.02-1.10), lactate dehydrogenase (OR, 1.002; 95% CI, 1.001-1.004) and direct bilirubin (OR, 1.15; 95% CI, 1.06-1.24). The mean AUC in the development cohort was 0.88 (95% CI, 0.85-0.91) and the AUC in the validation cohort was 0.88 (95% CI, 0.84-0.93). The score has been translated into an online risk calculator that is freely available to the public (http://118.126.104.170/). Conclusions and Relevance:In this study, a risk score based on characteristics of COVID-19 patients at the time of admission to the hospital was developed that may help predict a patient's risk of developing critical illness.
10.1001/jamainternmed.2020.2033
Nomogram-Based Prediction of the Risk of Diabetic Retinopathy: A Retrospective Study.
Journal of diabetes research
OBJECTIVES:This study is aimed at developing a risk nomogram of diabetic retinopathy (DR) in a Chinese population with type 2 diabetes mellitus (T2DM). METHODS:A questionnaire survey, biochemical indicator examination, and physical examination were performed on 4170 T2DM patients, and the collected data were used to evaluate the DR risk in T2DM patients. By operating R software, firstly, the least absolute shrinkage and selection operator (LASSO) regression analysis was used to optimize variable selection by running cyclic coordinate descent with 10 times cross-validation. Secondly, multivariable logistic regression analysis was applied to build a predicting model introducing the predictors selected from the LASSO regression analysis. The nomogram was developed based on the selected variables visually. Thirdly, calibration plot, receiver operating characteristic (ROC) curve, and decision curve analysis were used to validate the model, and further assessment was running by external validation. RESULTS:Seven predictors were selected by LASSO from 19 variables, including age, course of disease, postprandial blood glucose (PBG), glycosylated haemoglobin A1c (HbA1c), uric creatinine (UCR), urinary microalbumin (UMA), and systolic blood pressure (SBP). The model built by these 7 predictors displayed medium prediction ability with the area under the ROC curve of 0.700 in the training set and 0.715 in the validation set. The decision curve analysis curve showed that the nomogram could be applied clinically if the risk threshold is between 21% and 57% and 21%-51% in external validation. CONCLUSION:Introducing age, course of disease, PBG, HbA1c, UCR, UMA, and SBP, the risk nomogram is useful for prediction of DR risk in T2DM individuals.
10.1155/2020/7261047
The combined use of salivary biomarkers and clinical parameters to predict the outcome of scaling and root planing: A cohort study.
Liu Yiying,Duan Dingyu,Ma Rui,Ding Yi,Xu Yi,Zhou Xuedong,Zhao Lei,Xu Xin
Journal of clinical periodontology
AIM:To explore the application of the combined use of baseline salivary biomarkers and clinical parameters in predicting the outcome of scaling and root planing (SRP). MATERIALS AND METHODS:Forty patients with advanced periodontitis were included. Baseline saliva samples were analysed for interleukin-1β (IL-1β), matrix metalloproteinase-8 and the loads of Porphyromonas gingivalis, Prevotella intermedia, Aggregatibacter actinomycetemcomitans and Tannerella forsythia. After SRP, pocket closure and further attachment loss at 6 months post-treatment were chosen as outcome variables. Models to predict the outcomes were established by generalized estimating equations. RESULTS:The combined use of baseline clinical attachment level (CAL), site location and IL-1β (area under the curve [AUC] = 0.764) better predicted pocket closure than probing depth (AUC = 0.672), CAL (AUC = 0.679), site location (AUC = 0.654) or IL-1β (AUC = 0.579) alone. The combination of site location, tooth loss, percentage of deep pockets, detection of A. actinomycetemcomitans and T. forsythia load (AUC = 0.842) better predicted further clinical attachment loss than site location (AUC = 0.715), tooth loss (AUC = 0.530), percentage of deep pockets (AUC = 0.659) or T. forsythia load (AUC = 0.647) alone. CONCLUSION:The combination of baseline salivary biomarkers and clinical parameters better predicted SRP outcomes than each alone. The current study indicates the possible usefulness of salivary biomarkers in addition to tooth-related parameters in predicting SRP outcomes.
10.1111/jcpe.13367