Introduction: Leukocoria, a whitish pupillary reflection, is a common warning sign of various pediatric ocular pathologies, including Persistent hyperplastic primary vitreous (PHVP). This case report describes the obs...Introduction: Leukocoria, a whitish pupillary reflection, is a common warning sign of various pediatric ocular pathologies, including Persistent hyperplastic primary vitreous (PHVP). This case report describes the observation of a 3-year-old child with melanoderma, presenting with a white spot in the left eye since birth. Observation: Uncorrected distance visual acuity was 2/10 in the right eye and hand movements were perceived at 2 meters in the left eye. Examination of the anterior segment in the left eye revealed leukocoria and microphthalmia. The fundus examination was normal in the right eye but revealed a white mass extending from the center of the optic disc towards the temporal region in the left eye. Ocular imaging, including ocular ultrasound and optical coherence tomography, confirmed the diagnosis of PHVP, illustrated by a residual white mass at the center of the optic disc in the left eye. Conclusion: PHVP should be considered in the presence of leukocoria in a child, with urgent exclusion of retinoblastoma. This observation underscores the importance of early diagnosis for appropriate management. .展开更多
The advent of optical coherence tomography(OCT)imaging has changed the way ophthalmologists image the ocular surface and anterior segment of the eye.Its ability to obtain dynamic,high and ultra-high resolution,cross-s...The advent of optical coherence tomography(OCT)imaging has changed the way ophthalmologists image the ocular surface and anterior segment of the eye.Its ability to obtain dynamic,high and ultra-high resolution,cross-sectional images of the ocular surface and anterior segment in a noninvasive and rapid manner allows for ease of use.In this review,we focus on the use of anterior segment OCT,which provides an“optical biopsy”or in vivo imaging of various ocular surface and corneal pathologies,allowing the clinician to diagnose diseases otherwise not visualized by traditional methods.The utility of anterior segment OCT for various anterior segment pathologies is reviewed.展开更多
ackground:Uncorrected refractive error is a major cause of vision impairment worldwide and its increasing prevalent necessitates effective screening and management strategies.Meanwhile,deep learning,a subset of Artifi...ackground:Uncorrected refractive error is a major cause of vision impairment worldwide and its increasing prevalent necessitates effective screening and management strategies.Meanwhile,deep learning,a subset of Artificial Intelligence,has significantly advanced ophthalmological diagnostics by automating tasks that required extensive clinical expertise.Although recent studies have investigated the use of deep learning models for refractive power detection through various imaging techniques,a comprehensive systematic review on this topic is has yet be done.This review aims to summarise and evaluate the performance of ocular image-based deep learning models in predicting refractive errors.Main text:We search on three databases(PubMed,Scopus,Web of Science)up till June 2023,focusing on deep learning applications in detecting refractive error from ocular images.We included studies that had reported refractive error outcomes,regardless of publication years.We systematically extracted and evaluated the continuous outcomes(sphere,SE,cylinder)and categorical outcomes(myopia),ground truth measurements,ocular imaging modalities,deep learning models,and performance metrics,adhering to PRISMA guidelines.Nine studies were identified and categorised into three groups:retinal photo-based(n=5),OCT-based(n=1),and external ocular photo-based(n=3).For high myopia prediction,retinal photo-based models achieved AUC between 0.91 and 0.98,sensitivity levels between 85.10%and 97.80%,and specificity levels between 76.40%and 94.50%.For continuous prediction,retinal photo-based models reported MAE ranging from 0.31D to 2.19D,and R^(2) between 0.05 and 0.96.The OCT-based model achieved an AUC of 0.79–0.81,sensitivity of 82.30%and 87.20%and specificity of 61.70%–68.90%.For external ocular photo-based models,the AUC ranged from 0.91 to 0.99,sensitivity of 81.13%–84.00%and specificity of 74.00%–86.42%,MAE ranges from 0.07D to 0.18D and accuracy ranges from 81.60%to 96.70%.The reported papers collectively showed promising performances,in particular the retinal photo-based and external eye photo-based DL models.Conclusions:The integration of deep learning model and ocular imaging for refractive error detection appear promising.However,their real-world clinical utility in current screening workflow have yet been evaluated and would require thoughtful consideration in design and implementation.展开更多
Ptosis is a common ophthalmologic condition,and the diagnosis is primarily based on ocular appearance.Thediagnosis of such conditions can be improved using emerging technology such as artificial intelligence-basedmeth...Ptosis is a common ophthalmologic condition,and the diagnosis is primarily based on ocular appearance.Thediagnosis of such conditions can be improved using emerging technology such as artificial intelligence-basedmethods.However,unified data collection and labeling standards have not yet been established.This directlyimpacts the accuracy of ptosis diagnosis based on appearance alone.Therefore,in the present study,we aimedto establish a procedure to obtain and label images to devise a recommendation system for optimal recognitionof ptosis based on ocular appearances.This would help to standardize and facilitate data sharing and serve as aguideline for the development and improvisation of algorithms in artificial intelligence for ptosis.展开更多
Background:To evaluate the frequency and characteristics of sub-clinical ocular surface squamous neoplasia(OSSN)detected by high-resolution anterior segment tomography(HR-OCT)in patients with clinically unapparent dis...Background:To evaluate the frequency and characteristics of sub-clinical ocular surface squamous neoplasia(OSSN)detected by high-resolution anterior segment tomography(HR-OCT)in patients with clinically unapparent disease following topical treatment.Methods:A retrospective chart review of patients with OSSN identified through a pharmacy database at the Bascom Palmer Eye Institute from January 2013 to December 2018 was conducted.Patients undergoing primary therapy with topical 5-fluorouracil 1%(5-FU)(4 times a day for 7 days with a 21-day break)or interferon-alpha-2b(IFN)(4 times a day)were reviewed.Patients were separated into two groups.Group 1 included individuals whose clinical resolution of OSSN aligned with complete resolution on HR-OCT.Group 2(sub-clinical OSSN group)included individuals with clinical OSSN resolution but with features of persistent disease on HR-OCT.Patients excluded included those treated at an outside institution and those who used topical therapy as a surgical adjunct.Results:A total of 95 patients(95 eyes)were reviewed.Sub-clinical OSSN was detected at a frequency of 17%in our study patients(n=16 patients,9 treated with 5-FU and 7 treated with IFN).In the 16 individuals,the mean time to clinical resolution was 3.6±1.0 cycles for 5-FU and 4.0±0.0 months for IFN.An additional 2.1±0.8 cycles for 5-FU and 1.2±0.4 months for IFN were needed to achieve HR-OCT resolution of OSSN.Recurrence in Group 1 was noted in 10 patients(12%)while no recurrences occurred in Group 2,the cohort with subclinical disease that received the extended medical therapy.The mean follow-up was 24.0±17.9 months.Conclusion:We found that at least 17%of individuals with apparent clinical resolution of OSSN have sub-clinical disease detected on HR-OCT.This information can be used to optimize treatment and extend therapy past the point of clinical resolution.展开更多
Background The retinal image quality derived from lower-order(LOA)and higher-order aberrations(HOA)for fixed 3-mm and photopic pupil diameters,in children undergoing combined 0.01%atropine and orthokeratology(AOK)vers...Background The retinal image quality derived from lower-order(LOA)and higher-order aberrations(HOA)for fixed 3-mm and photopic pupil diameters,in children undergoing combined 0.01%atropine and orthokeratology(AOK)versus those receiving orthokeratology alone(OK)over two years was evaluated.Methods The visual Strehl ratio based on the optical transfer function(VSOTF),derived from 2nd-to 4th-order terms(LOA and HOA combined),2nd-order terms(LOA only),and 3rd-to 4th-order terms(HOA only)for fixed 3-mm and natural photopic pupil diameters,was compared between the two treatment groups.The individual Zernike coef-ficients fora fixed 3-mm pupil ize of 2nd to 4th,orders,root mean square(RMS)of LOA(Z_(2)^(0),Z_(2)^(-2),and Z_(2)^(2) combined),HOA(3rd to 4th orders inclusive),and coma(Z_(3)^(-1) and Z_(3)^(1) combined)were also compared between the two groups.Results Right eye data of 33 AOK and 35 OK participants were analysed.Under photopic conditions,significantly lower VSOTF based on HOA only was observed in the AOK group compared with that in the OK group at all post-treatment visits(all P<0.05);however,interactions between HOA and LOA resulted in comparable overall retinal image quality(i.e.,VSOTF based on LOA and HOA combined)between the two groups at all visits(all P>0.05).For a fixed 3-mm pupil size,the VSOTF based on HOA only,LOA only,or HOA and LOA combined,were not different between the two groups(all P>0.05).AOK participants had slower axial elongation(mean±SD,0.17±0.19 mm vs.0.35±0.20 mm,P<0.001),a larger photopic pupil size(4.05±0.61 mm vs.3.43±0.41 mm,P<0.001)than OK partici-pants overtwoyears.Conclusions HOA profile related to an enlarged pupil size may provide visual signal influencing eye growth in the AOK group.展开更多
Background:Conjunctival lymphoma,conjunctival amyloidosis and benign reactive lymphoid hyperplasia(BRLH)are conditions that often have a similar appearance on the ocular surface.The use of high resolution anterior seg...Background:Conjunctival lymphoma,conjunctival amyloidosis and benign reactive lymphoid hyperplasia(BRLH)are conditions that often have a similar appearance on the ocular surface.The use of high resolution anterior segment optical coherence tomography(HR-OCT)enables clinicians to evaluate distinctive differences in tissue morphology and cellular patterns in various ocular surface conditions.In this study,we characterize the morphological differences seen in conjunctival lymphoma,conjunctival amyloidosis and BRLH on HR-OCT imaging.Methods:A retrospective chart review was performed of patients with biopsy proven conjunctival lymphoma,conjunctival amyloidosis and BRLH between 2012 and 2019 at the Bascom Palmer Eye Institute.Patients were excluded if HR-OCT imaging was not performed on initial presentation.Results:Thirty-four total eyes of 27 patients were identified.Twenty eyes had conjunctival lymphoma(16 patients),8 eyes had conjunctival amyloidosis(6 patients)and 6 eyes had BRLH(5 patients).All conditions appeared clinically as pink,red or yellow subepithelial lesions but had different features on HR-OCT.In lymphoma,HR-OCT images typically showed homogenous,dark subepithelial lesions with smooth borders,containing monomorphic dot-like infiltrates.HR-OCT images of amyloidosis typically showed heterogeneous,dark lesions with irregular borders,often containing hyperreflective linear infiltrates.HR-OCT images of BRLH showed variable infiltration of the subepithelial tissue,at times with homogenous lesions containing dot-like infiltrates like lymphoma and other times with more hyperreflective,subepithelial tissue.Flow cytometry and gene rearrangement was needed for final differentiation between BRLH and lymphoma lesions.Conclusions:Distinctive features on HR-OCT of conjunctival lymphoma,conjunctival amyloidosis and BRLH can help characterize these lesions beyond what is apparent with the clinical examination.Future studies can further validate this technology’s use with more subtle and challenging lesions.展开更多
The global prevalence of diabetes is steadily increasing,with a high percentage of patients unaware of their disease status.Screening for diabetes is of great significance in preventive medicine and may benefit from d...The global prevalence of diabetes is steadily increasing,with a high percentage of patients unaware of their disease status.Screening for diabetes is of great significance in preventive medicine and may benefit from deep learning technology.In traditional Chinese medicine,specific features on the ocular surface have been explored as diagnostic indicators for systemic diseases.Here we explore the feasibility of using features from the entire ocular surface to construct deep learning models for risk assessment and detection of type 2 diabetes(T2DM).We performed an observational,multicenter study using ophthalmic images of the ocular surface to develop a deep convolutional network,OcularSurfaceNet.The deep learning system was trained and validated with a multicenter dataset of 416580 images from 67151 participants and tested independently using an additional 91422 images from 12544 participants,and can be used to identify individuals at high risk of T2DM with areas under the receiver operating characteristic curve(AUROC)of 0.89-0.92 and T2DM with AUROC of 0.70-0.82.Our study demonstrated a qualitative relationship between ocular surface images and T2DM risk level,which provided new insights for the potential utility of ocular surface images in T2DM screening.Overall,our findings suggest that the deep learning framework using ocular surface images can serve as an opportunistic screening toolkit for noninvasive and low-cost large-scale screening of the general population in risk assessment and early identification of T2DM patients.展开更多
文摘Introduction: Leukocoria, a whitish pupillary reflection, is a common warning sign of various pediatric ocular pathologies, including Persistent hyperplastic primary vitreous (PHVP). This case report describes the observation of a 3-year-old child with melanoderma, presenting with a white spot in the left eye since birth. Observation: Uncorrected distance visual acuity was 2/10 in the right eye and hand movements were perceived at 2 meters in the left eye. Examination of the anterior segment in the left eye revealed leukocoria and microphthalmia. The fundus examination was normal in the right eye but revealed a white mass extending from the center of the optic disc towards the temporal region in the left eye. Ocular imaging, including ocular ultrasound and optical coherence tomography, confirmed the diagnosis of PHVP, illustrated by a residual white mass at the center of the optic disc in the left eye. Conclusion: PHVP should be considered in the presence of leukocoria in a child, with urgent exclusion of retinoblastoma. This observation underscores the importance of early diagnosis for appropriate management. .
基金Ronald and Alicia Lepke Grant,The Lee and Claire Hager Grant,The Jimmy and Gaye Bryan Grant,The H.Scott Huizenga Grant,The Grant and Diana Stanton-Thornbrough,The Robert Baer Family Grant,The Emilyn Page and Mark Feldberg Grant,The Gordon Charitable Foundation,The Richard and Kathy Lesser Grant and The Richard Azar Family Grant(institutional grants).
文摘The advent of optical coherence tomography(OCT)imaging has changed the way ophthalmologists image the ocular surface and anterior segment of the eye.Its ability to obtain dynamic,high and ultra-high resolution,cross-sectional images of the ocular surface and anterior segment in a noninvasive and rapid manner allows for ease of use.In this review,we focus on the use of anterior segment OCT,which provides an“optical biopsy”or in vivo imaging of various ocular surface and corneal pathologies,allowing the clinician to diagnose diseases otherwise not visualized by traditional methods.The utility of anterior segment OCT for various anterior segment pathologies is reviewed.
文摘ackground:Uncorrected refractive error is a major cause of vision impairment worldwide and its increasing prevalent necessitates effective screening and management strategies.Meanwhile,deep learning,a subset of Artificial Intelligence,has significantly advanced ophthalmological diagnostics by automating tasks that required extensive clinical expertise.Although recent studies have investigated the use of deep learning models for refractive power detection through various imaging techniques,a comprehensive systematic review on this topic is has yet be done.This review aims to summarise and evaluate the performance of ocular image-based deep learning models in predicting refractive errors.Main text:We search on three databases(PubMed,Scopus,Web of Science)up till June 2023,focusing on deep learning applications in detecting refractive error from ocular images.We included studies that had reported refractive error outcomes,regardless of publication years.We systematically extracted and evaluated the continuous outcomes(sphere,SE,cylinder)and categorical outcomes(myopia),ground truth measurements,ocular imaging modalities,deep learning models,and performance metrics,adhering to PRISMA guidelines.Nine studies were identified and categorised into three groups:retinal photo-based(n=5),OCT-based(n=1),and external ocular photo-based(n=3).For high myopia prediction,retinal photo-based models achieved AUC between 0.91 and 0.98,sensitivity levels between 85.10%and 97.80%,and specificity levels between 76.40%and 94.50%.For continuous prediction,retinal photo-based models reported MAE ranging from 0.31D to 2.19D,and R^(2) between 0.05 and 0.96.The OCT-based model achieved an AUC of 0.79–0.81,sensitivity of 82.30%and 87.20%and specificity of 61.70%–68.90%.For external ocular photo-based models,the AUC ranged from 0.91 to 0.99,sensitivity of 81.13%–84.00%and specificity of 74.00%–86.42%,MAE ranges from 0.07D to 0.18D and accuracy ranges from 81.60%to 96.70%.The reported papers collectively showed promising performances,in particular the retinal photo-based and external eye photo-based DL models.Conclusions:The integration of deep learning model and ocular imaging for refractive error detection appear promising.However,their real-world clinical utility in current screening workflow have yet been evaluated and would require thoughtful consideration in design and implementation.
基金The study was supported by Science and Technology PlanningProjects of Guangdong Province(Grant No.2018B010109008)National Key R&D Program of China(Grant No.2018YFC0116500).
文摘Ptosis is a common ophthalmologic condition,and the diagnosis is primarily based on ocular appearance.Thediagnosis of such conditions can be improved using emerging technology such as artificial intelligence-basedmethods.However,unified data collection and labeling standards have not yet been established.This directlyimpacts the accuracy of ptosis diagnosis based on appearance alone.Therefore,in the present study,we aimedto establish a procedure to obtain and label images to devise a recommendation system for optimal recognitionof ptosis based on ocular appearances.This would help to standardize and facilitate data sharing and serve as aguideline for the development and improvisation of algorithms in artificial intelligence for ptosis.
基金NIH Center Core Grant P30EY014801Research to Prevent Blindness+10 种基金Department of Veterans AffairsVeterans Health AdministrationOffice of Research and DevelopmentClinical Sciences Research EPID-006-15S(Dr.Galor)R01EY026174(Dr.Galor)The Dr.Ronald and Alicia Lepke Grant,The Lee and Claire Hager Grant,The Jimmy and Gaye Bryan Grant,The H.Scott Huizenga Grant,The Grant and Diana Stanton-ThornbroughThe Robert Baer Family GrantThe Emilyn Page and Mark Feldberg GrantThe Jose Ferreira de Melo Grant,Richard and Kathy Lesser GrantThe Michele and Ted Kaplan Grantthe Richard Azar Family Grant(institutional grants).
文摘Background:To evaluate the frequency and characteristics of sub-clinical ocular surface squamous neoplasia(OSSN)detected by high-resolution anterior segment tomography(HR-OCT)in patients with clinically unapparent disease following topical treatment.Methods:A retrospective chart review of patients with OSSN identified through a pharmacy database at the Bascom Palmer Eye Institute from January 2013 to December 2018 was conducted.Patients undergoing primary therapy with topical 5-fluorouracil 1%(5-FU)(4 times a day for 7 days with a 21-day break)or interferon-alpha-2b(IFN)(4 times a day)were reviewed.Patients were separated into two groups.Group 1 included individuals whose clinical resolution of OSSN aligned with complete resolution on HR-OCT.Group 2(sub-clinical OSSN group)included individuals with clinical OSSN resolution but with features of persistent disease on HR-OCT.Patients excluded included those treated at an outside institution and those who used topical therapy as a surgical adjunct.Results:A total of 95 patients(95 eyes)were reviewed.Sub-clinical OSSN was detected at a frequency of 17%in our study patients(n=16 patients,9 treated with 5-FU and 7 treated with IFN).In the 16 individuals,the mean time to clinical resolution was 3.6±1.0 cycles for 5-FU and 4.0±0.0 months for IFN.An additional 2.1±0.8 cycles for 5-FU and 1.2±0.4 months for IFN were needed to achieve HR-OCT resolution of OSSN.Recurrence in Group 1 was noted in 10 patients(12%)while no recurrences occurred in Group 2,the cohort with subclinical disease that received the extended medical therapy.The mean follow-up was 24.0±17.9 months.Conclusion:We found that at least 17%of individuals with apparent clinical resolution of OSSN have sub-clinical disease detected on HR-OCT.This information can be used to optimize treatment and extend therapy past the point of clinical resolution.
基金supported by The Hong Kong Polytechnic University Teaching Postgraduate Studentship(TPS)Scheme to QiTansupported by a private donation to the School of Optometry,The Hong Kong Polytechnic University.
文摘Background The retinal image quality derived from lower-order(LOA)and higher-order aberrations(HOA)for fixed 3-mm and photopic pupil diameters,in children undergoing combined 0.01%atropine and orthokeratology(AOK)versus those receiving orthokeratology alone(OK)over two years was evaluated.Methods The visual Strehl ratio based on the optical transfer function(VSOTF),derived from 2nd-to 4th-order terms(LOA and HOA combined),2nd-order terms(LOA only),and 3rd-to 4th-order terms(HOA only)for fixed 3-mm and natural photopic pupil diameters,was compared between the two treatment groups.The individual Zernike coef-ficients fora fixed 3-mm pupil ize of 2nd to 4th,orders,root mean square(RMS)of LOA(Z_(2)^(0),Z_(2)^(-2),and Z_(2)^(2) combined),HOA(3rd to 4th orders inclusive),and coma(Z_(3)^(-1) and Z_(3)^(1) combined)were also compared between the two groups.Results Right eye data of 33 AOK and 35 OK participants were analysed.Under photopic conditions,significantly lower VSOTF based on HOA only was observed in the AOK group compared with that in the OK group at all post-treatment visits(all P<0.05);however,interactions between HOA and LOA resulted in comparable overall retinal image quality(i.e.,VSOTF based on LOA and HOA combined)between the two groups at all visits(all P>0.05).For a fixed 3-mm pupil size,the VSOTF based on HOA only,LOA only,or HOA and LOA combined,were not different between the two groups(all P>0.05).AOK participants had slower axial elongation(mean±SD,0.17±0.19 mm vs.0.35±0.20 mm,P<0.001),a larger photopic pupil size(4.05±0.61 mm vs.3.43±0.41 mm,P<0.001)than OK partici-pants overtwoyears.Conclusions HOA profile related to an enlarged pupil size may provide visual signal influencing eye growth in the AOK group.
基金Ronald and Alicia Lepke Grant,The Lee and Claire Hager Grant,The Jimmy and Gaye Bryan Grant,The H.Scott Huizenga Grant,The Grant and Diana Stanton-Thornbrough,The Robert Baer Family Grant,The Emilyn Page and Mark Feldberg Grant,The Gordon Charitable Foundation,The Jose Ferreira de Melo Grant,The Richard and Kathy Lesser Grant and The Richard Azar Family Grant(institutional grants),the Department of Veterans Affairs,Veterans Health Administration,Office of Research and Development,Clinical Sciences Research EPID-006-15S(Dr.Galor),R01EY026174(Dr.Galor)NIH Center Core Grant P30EY014801Research to Prevent Blindness Unrestricted Grant.
文摘Background:Conjunctival lymphoma,conjunctival amyloidosis and benign reactive lymphoid hyperplasia(BRLH)are conditions that often have a similar appearance on the ocular surface.The use of high resolution anterior segment optical coherence tomography(HR-OCT)enables clinicians to evaluate distinctive differences in tissue morphology and cellular patterns in various ocular surface conditions.In this study,we characterize the morphological differences seen in conjunctival lymphoma,conjunctival amyloidosis and BRLH on HR-OCT imaging.Methods:A retrospective chart review was performed of patients with biopsy proven conjunctival lymphoma,conjunctival amyloidosis and BRLH between 2012 and 2019 at the Bascom Palmer Eye Institute.Patients were excluded if HR-OCT imaging was not performed on initial presentation.Results:Thirty-four total eyes of 27 patients were identified.Twenty eyes had conjunctival lymphoma(16 patients),8 eyes had conjunctival amyloidosis(6 patients)and 6 eyes had BRLH(5 patients).All conditions appeared clinically as pink,red or yellow subepithelial lesions but had different features on HR-OCT.In lymphoma,HR-OCT images typically showed homogenous,dark subepithelial lesions with smooth borders,containing monomorphic dot-like infiltrates.HR-OCT images of amyloidosis typically showed heterogeneous,dark lesions with irregular borders,often containing hyperreflective linear infiltrates.HR-OCT images of BRLH showed variable infiltration of the subepithelial tissue,at times with homogenous lesions containing dot-like infiltrates like lymphoma and other times with more hyperreflective,subepithelial tissue.Flow cytometry and gene rearrangement was needed for final differentiation between BRLH and lymphoma lesions.Conclusions:Distinctive features on HR-OCT of conjunctival lymphoma,conjunctival amyloidosis and BRLH can help characterize these lesions beyond what is apparent with the clinical examination.Future studies can further validate this technology’s use with more subtle and challenging lesions.
基金supported by the Science and Technology Planning Project of Inner Mongolia Autonomous Region[201802146]National Key Research and Development Program of China[2018YFC1707601,2018YFC1704200]+1 种基金Major Basic and Applied Basic Research Projects of Guangdong Province of China[2019B030302005]Mongolian Medicine Standardization Project of Inner Mongolia Autonomous Region People's Government[2018001].
文摘The global prevalence of diabetes is steadily increasing,with a high percentage of patients unaware of their disease status.Screening for diabetes is of great significance in preventive medicine and may benefit from deep learning technology.In traditional Chinese medicine,specific features on the ocular surface have been explored as diagnostic indicators for systemic diseases.Here we explore the feasibility of using features from the entire ocular surface to construct deep learning models for risk assessment and detection of type 2 diabetes(T2DM).We performed an observational,multicenter study using ophthalmic images of the ocular surface to develop a deep convolutional network,OcularSurfaceNet.The deep learning system was trained and validated with a multicenter dataset of 416580 images from 67151 participants and tested independently using an additional 91422 images from 12544 participants,and can be used to identify individuals at high risk of T2DM with areas under the receiver operating characteristic curve(AUROC)of 0.89-0.92 and T2DM with AUROC of 0.70-0.82.Our study demonstrated a qualitative relationship between ocular surface images and T2DM risk level,which provided new insights for the potential utility of ocular surface images in T2DM screening.Overall,our findings suggest that the deep learning framework using ocular surface images can serve as an opportunistic screening toolkit for noninvasive and low-cost large-scale screening of the general population in risk assessment and early identification of T2DM patients.
