A multi-centre prospective evaluation of THEIA™ to detect diabetic retinopathy (DR) and diabetic macular oedema (DMO) in the New Zealand screening program

Abstract

Purpose To validate the potential application of THEIA™ as clinical decision making assistant in a national screening program. Methods A total of 900 patients were recruited from either an urban large eye hospital, or a semi-rural optometrist led screening provider, as they were attending their appointment as part of New Zealand Diabetic Eye Screening Programme. The de-identified images were independently graded by three senior specialists, and final results were aggregated using New Zealand grading scheme, which was then converted to referable/non-referable and Healthy/mild/more than mild/sight threatening categories. Results THEIA™ managed to grade all images obtained during the study. Comparing the adjudicated images from the specialist grading team, “ground truth”, with the grading by the AI platform in detecting “sight threatening” disease, at the patient level THEIA™ achieved 100% imageability, 100% [98.49–100.00%] sensitivity and [97.02–99.16%] specificity, and negative predictive value of 100%. In other words, THEIA™ did not miss any patients with “more than mild” or “sight threatening” disease. The level of agreement between the clinicians and the aggregated results was (k value: 0.9881, 0.9557, and 0.9175), and the level of agreement between THEIA™ and the aggregated labels was (k value: 0.9515). Conclusion This multi-centre prospective trial showed that THEIA™ did not miss referable disease when screening for diabetic retinopathy and maculopathy. It also had a very high level of granularity in reporting the disease level. As THEIA™ has been tested on a variety of cameras, operating in a range of clinics (rural/urban, ophthalmologist-led\optometrist-led), we believe that it will be a suitable addition to a public diabetic screening program.

Use of artificial intelligence on retinal images to accurately predict the risk of cardiovascular event (CVD-AI)

“…Previously we have demonstrated that it is possible to train an artificial intelligence (AI) deep learning (DL) algorithm on retinal images to grade diabetic retinopathy and maculopathy for diagnostic, screening and risk assessment purposes [30,[36][37][38][39][40][41][42][43][44][45]. In this study we used 110,272 fundus images from a database of 55,118 patients from the UK Biobank and AREDS 1 datasets to train and subsequently test a novel AI platform (CVD-AI) to calculate a 10-year CVD risk score for these individuals.…”

Development and validation of a deep-learning model to predict 10-year atherosclerotic cardiovascular disease risk from retinal images using the UK Biobank and EyePACS 10K datasets

“…The retinopathy, maculopathy, drusen, pigmentary abnormality, advanced age-related macular degeneration, and smoking CNNs were previously trained on other datasets. 16 , 22 , 23 The rest of the CNNs were trained using the unique UK Biobank labels in the fundus images: “hba1c_result,” “tchdl_result,” “systolic_bp,” “systolic_bp2,” “smoking_status.” “Systolic_bp” and “systolic_bp2” are 2 consecutive blood pressure measurements in the UK Biobank and in this study we used the mean of the two.…”

“…This model was built upon our previous work on detecting and grading retinopathy, maculopathy, macular degeneration, and effects of smoking in retinal images. 16 , 17 , 18 Although the Framingham risk score is recommended to perform cardiovascular risk assessment in some countries, 19 the 2018 American Heart Association (AHA) Cholesterol Clinical Practice Guidelines recommend using the US-derived PCE to estimate the 10-year risk for hard ASCVD events (coronary heart disease death, nonfatal myocardial infarction, fatal or nonfatal stroke). 15 Our DL model was trained on and validated against the 10-year ASCVD risk as calculated by the PCE from individuals in the UK Biobank dataset (Level 3 in Figure 1 ; Supplemental Appendix 2 ).…”

Examination of alternative eGFR definitions on the performance of deep learning models for detection of chronic kidney disease from fundus photographs

“…175,788 fundus images from 85,707 individuals were obtained from the UK Biobank. A DL image screening system similar to the one used in our previous study [ 28 ] was created to screen for poor-quality images. In brief, a random set of images were first sampled from the UK Biobank dataset.…”