Diagnostic performance of deep-learning-based screening methods for diabetic retinopathy in primary care—A meta-analysis

Abstract

Background Diabetic retinopathy (DR) affects 10–24% of patients with diabetes mellitus type 1 or 2 in the primary care (PC) sector. As early detection is crucial for treatment, deep learning screening methods in PC setting could potentially aid in an accurate and timely diagnosis. Purpose The purpose of this meta-analysis was to determine the current state of knowledge regarding deep learning (DL) screening methods for DR in PC. Data sources A systematic literature search was conducted using Medline, Web of Science, and Scopus to identify suitable studies. Study selection Suitable studies were selected by two researchers independently. Studies assessing DL methods and the suitability of these screening systems (diagnostic parameters such as sensitivity and specificity, information on datasets and setting) in PC were selected. Excluded were studies focusing on lesions, applying conventional diagnostic imaging tools, conducted in secondary or tertiary care, and all publication types other than original research studies on human subjects. Data extraction The following data was extracted from included studies: authors, title, year of publication, objectives, participants, setting, type of intervention/method, reference standard, grading scale, outcome measures, dataset, risk of bias, and performance measures. Data synthesis and conclusion The summed sensitivity of all included studies was 87% and specificity was 90%. Given a prevalence of DR of 10% in patients with DM Type 2 in PC, the negative predictive value is 98% while the positive predictive value is 49%. Limitations Selected studies showed a high variation in sample size and quality and quantity of available data.

Determinants of the implementation of an artificial intelligence-supported device for the screening of diabetic retinopathy in primary care – a qualitative study

“…This, however, might lead to more false-positive results which need to be re-examined by the ophthalmologist and create an unnecessary feeling of insecurity for patients due to a false-positive pre-diagnosis. 19 Another strategy could be to emphasize the point that the use of AI is a question of indication. In the case of a screening situation, AI might constitute a comfortable, cost-and time-efficient solution for a pre-selection of patients.…”

“…This, however, might lead to more false-positive results which need to be re-examined by the ophthalmologist and create an unnecessary feeling of insecurity for patients due to a false-positive pre-diagnosis. 19…”

“…The guided questionnaire was developed by experiences of the researchers (postgraduate trainee, economic engineer and GP), results of a metaanalysis 19 within a project addressing AI and diagnosing DR 31 and brainstorming. It was piloted prior to the interviews with three GPs working at the Institute of Family Medicine in Lübeck.…”

“…[16][17][18] DL takes up a long history of autonomous and assisted analysis of retinal photographs that has existed since the 1990s and before. A metaanalysis from 2020, including 10 studies of DR screening in primary care, attributed a high diagnostic performance to DLalgorithms, with a pooled sensitivity of 87% and pooled specificity of 90% for referable DR. 19 A study from 2019 even revealed that in comparison with human graders, the DL algorithm had a significantly higher sensitivity (97% vs. 74%) in detecting referable DR. Also, the algorithms significantly reduced the false negative rate (by 23%) at the cost of slightly higher false positive rates (2%). 20 Especially in a screening situation, the sensitivity and prevalence are important because it determines the degree of false negative results.…”

Luminosity Normalized Symmetric Deep Convolute Tubular Classifier for Diabetic Retinopathy Detection

“…A meta-analysis study for DR in primary care was analyzed in [24] by employing deep learning-based screening methods. However, the negative impact of background information on classification results was not analyzed.…”