Recent findings suggest that with more research, abnormality detection may be an option for general retinal disease screening and identification of new presentations of common retinal diseases.

Abnormality detection may be useful in diabetic retinopathy screening for retinal diseases and in more general settings, suggest findings published in JAMA Ophthalmology. Anomaly detectors could potentially screen general populations for retinal diseases, identify new diseases, and phenotype or detect lesser-known manifestations of common retinal diseases.

Typically, the algorithms of a deep learning system (DLS), a type of artificial intelligence (AI), are trained to diagnose specific diseases based on available sample images. But rare presentations of common retinal diseases or rare ophthalmic conditions have limited annotated retinal images available, making it difficult for DLS training to recognize their respective features and accurately diagnose them.

Anomaly detection, on the other hand, can potentially be trained based only on normal images of retinas to then identify any new or abnormal presentation that may indicate disease or identify new disease phenotypes. This could be done without using images of abnormal retinas to train the algorithm. This AI could then be applied in various applications, including screening the general population, rather than being used only to identify individual conditions such as diabetic retinopathy, for example.

The researchers evaluated 16 anomaly detectors using 88,692 retinal images from 44,346 people with varying degrees of diabetic retinopathy. Images were extracted from the publicly available EyePACS dataset with fundus images labeled from 0 to 4 for retinopathy severity. There are 2 categories of DLS, both of which were included in the study. Discriminative networks classify and segment, which makes them useful in a task such as anomaly detection. Generative networks generate realistic synthetic data, such as backgrounds showing diabetic retinopathy.

The anomaly detectors were tested with a surrogate problem: they were trained with diabetic retinopathy images showing only retinas with non-referable diabetic retinopathy and no diabetic macular edema, no diabetic retinopathy or mild non-proliferative diabetic retinopathy to moderate. But non-referable and referable diabetic retinopathy – including diabetic macular edema or non-proliferative diabetic retinopathy – have been used to test the system’s recognition of retinal disease. Performance measures, including area under the receiver operating characteristic curve, F1 score, and accuracy, were used to evaluate the anomaly detectors.

A combination of Deep InfoMax for integration and a 1-class support vector machine for anomaly detection resulted in an area under the receiver operating characteristic of 0.808 (95% CI, 0.789-0.827 ) – the best performance on all systems in the study. It was a discriminating system, 73.9% accurate, and had an F1 score of 0.755. Descriptive systems also performed best overall. In the generative category, the InfoStyleGAN integration performed best, with an area of ​​0.667 under the receiver operating characteristic.

Overall, the results suggest that even in the absence of abnormality-specific data, abnormality detection was a useful tool – particularly in the context of diabetic retinopathy screening for retinal disease, in this case.

“When abnormal (diseased) data, i.e., referable diabetic retinopathy in this study, were not available for training retinal diagnostic systems in which only non-referable diabetic retinopathy was used for training , anomaly detection techniques were useful in identifying images with and without referable diabetic retinopathy,” the authors wrote.

Limitations of the study included the use of the limited surrogate problem and the availability of data only from the EyePACS system. However, the study authors believe there is potential for more widespread use with further research.

“Anomaly detectors can be used to detect retinal disease in more generalized settings and could potentially play a role in screening populations for retinal disease or identifying new diseases and phenotyping or detecting unusual presentations. of common retinal diseases,” they concluded.

Reference

Burlina P, Paul W, Liu TYA, Bressler NM. Detecting Abnormalities in Retinal Diseases Using Generative, Discriminative, and Self-supervised Deep. JAMA Ophthalmol. Published online December 30, 2021. doi:10.1001/jamaophthalmol.2021.5557