A deep learning algorithm to detect chronic kidney disease from retinal photographs in community-based populations

Sabanayagam, Charumathi; Xu, Dejiang; Ting, Daniel Shu Wei; Nusinovici, Simon; Banu, Riswana; Hamzah, Haslina; Lim, Cynthia Ciwei; Tham, Yih‐Chung; Cheung, Carol Y.; Tai, E. Shyong; Wang, Ya Xing; Jonas, Jost B.; Cheng, Ching-Yu; Lee, Mong Li; Hsu, Wynne; Wong, Tien Yin

doi:10.1016/s2589-7500(20)30063-7

Cited by 169 publications

(144 citation statements)

References 29 publications

Supporting

Mentioning

138

Contrasting

Order By: Relevance

“…Some possible explanations for this performance include more profound microvascular damage in patients with worse glucose control and the coexistence of signs of diabetic retinopathy and diabetic nephropathy, which were noted to be significantly associated [ 31 , 32 ]. A deep learning algorithm was recently formulated by a research group at the Singapore National Eye Center (SNEC) [ 33 ]. Their algorithm was used to detect CKD with eGFR <60 mL/min/1.73 m 2 by using both retinal images and risk factors, individually and in combination, in 3 population-based screening databases from Singapore and China [ 33 ].…”

Section: Discussionmentioning

confidence: 99%

Deep Learning–Based Detection of Early Renal Function Impairment Using Retinal Fundus Images: Model Development and Validation

Kang¹,

Hsieh²,

Li³

et al. 2020

JMIR Med Inform

View full text Add to dashboard Cite

Background Retinal imaging has been applied for detecting eye diseases and cardiovascular risks using deep learning–based methods. Furthermore, retinal microvascular and structural changes were found in renal function impairments. However, a deep learning–based method using retinal images for detecting early renal function impairment has not yet been well studied. Objective This study aimed to develop and evaluate a deep learning model for detecting early renal function impairment using retinal fundus images. Methods This retrospective study enrolled patients who underwent renal function tests with color fundus images captured at any time between January 1, 2001, and August 31, 2019. A deep learning model was constructed to detect impaired renal function from the images. Early renal function impairment was defined as estimated glomerular filtration rate <90 mL/min/1.73 m2. Model performance was evaluated with respect to the receiver operating characteristic curve and area under the curve (AUC). Results In total, 25,706 retinal fundus images were obtained from 6212 patients for the study period. The images were divided at an 8:1:1 ratio. The training, validation, and testing data sets respectively contained 20,787, 2189, and 2730 images from 4970, 621, and 621 patients. There were 10,686 and 15,020 images determined to indicate normal and impaired renal function, respectively. The AUC of the model was 0.81 in the overall population. In subgroups stratified by serum hemoglobin A1c (HbA1c) level, the AUCs were 0.81, 0.84, 0.85, and 0.87 for the HbA1c levels of ≤6.5%, >6.5%, >7.5%, and >10%, respectively. Conclusions The deep learning model in this study enables the detection of early renal function impairment using retinal fundus images. The model was more accurate for patients with elevated serum HbA1c levels.

show abstract

Section: Discussionmentioning

confidence: 99%

Deep Learning–Based Detection of Early Renal Function Impairment Using Retinal Fundus Images: Model Development and Validation

Kang¹,

Hsieh²,

Li³

et al. 2020

JMIR Med Inform

View full text Add to dashboard Cite

show abstract

“…Figure 1 : Manhattan plot of genome-wide association study for retinal vessel tortuosity corrected for phenotypic variables that showed a statistically significant association, i.e. age, sex, and a subset of principal components of genotypes (PCs: 1,2, 5,6,7,8,16,17,18). Refer to Supplementary Material analysis for correlation with potential confounders.…”

Section: Shared Genetic Architecture With Diseasementioning

confidence: 99%

GWAS of Retinal Vessel Tortuosity Identifies 173 Novel Loci Revealing Genes and Pathways Associated with Vascular Pathomechanics and Cardiometabolic Diseases

Tomasoni

Mounier

Porcu

et al. 2020

Preprint

View full text Add to dashboard Cite

Fundus pictures of the eye allow for non-invasive inspection of the microvasculature system of the retina which is informative on cardiovascular health. Automated image processing enables the extraction of morphometric properties of this system as quantitative features that can be used for modelling disease risks. Here we report the results of the largest genome-wide association study (GWAS) of retinal vessel tortuosity conducted to date using data from the UK Biobank (N=63,899). We identified 87 loci associated with this trait (85 of which are novel). The heritability of the trait was h2=0.23 (0.02). We carried out a replication study on a small independent population-based cohort, SKIPOGH (N=436). While the power of this study was too small to replicate individual hits, the effect size estimates correlated significantly between the two studies (Pearson correlation r =0.55, p=4.6E-6). Using LD score regression, we showed that the alleles associated with retinal vessel tortuosity point to a common genetic architecture of this trait with CVD and related traits. Our results shed new light on the genetics of cardiovascular risk factors and disease.

show abstract

“…Moreover, CKD awareness is as low as 10% and studies to identify the potential benefits and risks of screening, screening measures, and target groups for screening of asymptomatic individuals, would help inform inconsistent screening guidelines that exist across professional bodies [52]. Advances in retinal fundus imaging technology and integration with machine learning approaches will enable rapid, non-invasive, point-of-care diagnoses, which may enhance screening service provision and improved screening compliance [53]. Retinal cameras are common in primary care settings and high street opticians for diabetic retinopathy screening.…”

Section: Discussionmentioning

confidence: 99%

“…Retinal cameras are common in primary care settings and high street opticians for diabetic retinopathy screening. Recent advances in smartphone technology, combined with the utility of machine learning approaches, highlights the feasibility and potential offered by non-invasive retinal photography as an adjunctive or opportunistic screening tool for CKD in the community [53].…”

Section: Discussionmentioning

confidence: 99%

Association of retinal venular tortuosity with impaired renal function in the Northern Ireland Cohort for the Longitudinal Study of Ageing

et al. 2020

View full text Add to dashboard Cite

Background: Previous studies have identified retinal microvascular features associated with renal dysfunction. Biopsies are necessary to confirm kidney microvascular damage and retinal imaging may enable evaluation of microangiopathic characteristics reflecting renal changes associated with chronic kidney disease (CKD). We evaluated retinal microvascular parameters (RMPs) for associations with renal function in a cross-sectional analysis of the Northern Ireland Cohort for the Longitudinal Study of Ageing. Methods: RMPs (central retinal arteriolar/ venular equivalents [CRAE/CRVE], arteriolar to venular ratio [AVR], fractal dimension and tortuosity) were measured from optic disc centred fundus images using semi-automated software. Associations were assessed with multivariable regression analyses between RMPs and estimated glomerular filtration rate (eGFR) defined by serum creatinine (eGFRscr) and cystatin C (eGFRcys) and also CKD status characterised by eGFR < 60 mL/min/1.73m 2. Regression models were adjusted for potential confounders including age, sex, diabetes, smoking status, educational attainment, cardiovascular disease, body mass index, antihypertensive medication, systolic blood pressure, triglycerides, high-and low-density lipoprotein levels. Results: Data were included for 1860 participants that had measures of renal function and retinal fundus images of sufficient quality for analysis. Participants had a mean age of 62.0 ± 8.5 yrs. and 53% were female. The mean eGFR for scr and cys were 82.2 ± 14.9 mL/min/1.73m 2 and 70.7 ± 18.6 mL/min/1.73m 2 respectively. eGFRcys provided lower estimates than eGFRscr resulting in a greater proportion of participants categorised as having CKD stages 3-5 (eGFRcys 26.8%; eGFRscr 7.9%). Multivariable regression analyses showed that increased venular tortuosity (OR = 1.30; 95%CI: 1.10, 1.54; P < 0.01) was associated with CKD stages 3-5 characterised by eGFRscr < 60 mL/min/1.73 m 2. No additional associations between CKD status characterised by eGFRscr or with eGFRcys, were detected (P > 0.05). Multivariable regression failed to detect associations between CRAE, CRVE, AVR, fractal dimension or tortuosity and eGFRscr or eGFRcys (P > 0.05). Conclusion: Increased retinal venular tortuosity was associated with CKD stages 3-5 defined by eGFRscr < 60 mL/ min/1.73 m 2 , in an older population independent of potential confounding factors. These retinal measures may provide non-invasive microvascular assessment of associations with CKD.

show abstract

A deep learning algorithm to detect chronic kidney disease from retinal photographs in community-based populations

Cited by 169 publications

References 29 publications

Deep Learning–Based Detection of Early Renal Function Impairment Using Retinal Fundus Images: Model Development and Validation

Deep Learning–Based Detection of Early Renal Function Impairment Using Retinal Fundus Images: Model Development and Validation

GWAS of Retinal Vessel Tortuosity Identifies 173 Novel Loci Revealing Genes and Pathways Associated with Vascular Pathomechanics and Cardiometabolic Diseases

Association of retinal venular tortuosity with impaired renal function in the Northern Ireland Cohort for the Longitudinal Study of Ageing

Contact Info

Product

Resources

About