Artificial intelligence (AI) is increasingly entering our daily lives and, even in the ophthalmic field, potential applications are being investigated. In particular, this article will examine the applications of AI in optical coherence tomography (OCT) in diabetic macular oedema.
Diabetic macular oedema (EMD) is a serious ocular complication of diabetes mellitus, both type 1 and type 2, and is one of the leading causes of vision loss in developed countries. Currently, approximately 537 million adults are living with diabetes and this number is expected to increase steadily as the average age of the population increases. The prevalence of EMD among people with diabetes in Europe has been estimated at 3.7% and its mean annual aggregate incidence in patients with type 2 diabetes is 0.4%.Â
In recent years, significant advances have been made in telecommunication, artificial intelligence (AI) and deep learning (DL) systems technologies, which have opened up new horizons for the creation of efficient tools for the quantification of key parameters relevant to the diagnosis and monitoring of macular disorders. Data from the literature suggest that AI can achieve high performance in detecting retinal features and assessing anatomical changes characteristic of diabetic macular oedema. Furthermore, artificial intelligence has already been successfully used to screen for diabetic retinopathy, using photographs of the ocular fundus, albeit with variable protocols, enabling early diagnosis, resulting in avoidable blindness and saving on disease management costs.Â
Diagnosis of diabetic macular oedema: the state of the art
EMD is a complex multifactorial disease, characterised by hypoxia, inflammation, hyperpermeability and angiogenesis. Consequently, patients may present different phenotypes, with important differences in disease severity, risk of progression and treatment outcomes. Therefore, the assessment of individual morphological features of EMD may provide a better understanding of the pathophysiology of this disease, which, in turn, may help in the selection of the best therapeutic option and a personalised medical approach.
In this context, the optical coherence tomography (OCT), and in particular Spectral-Domain OCT (SD-OCT), opens up new perspectives in terms of a significant improvement in diagnostic efficacy, with the possibility of obtaining quantitative and qualitative information of the morphological characteristics of the inner and outer retina. Among the main biomarkers identified by OCT for the diagnosis of EMD are the presence and amount of intraretinal (IRF) and subretinal fluid (SRF), the integrity of the outer limiting membrane (ELM) and ellipsoid zone (EZ), and the number of hyperreflective foci (HRF).
AI as a reliable tool to detect and quantify different OCT biomarkers in DME: a study
A recent study The aim was to validate an AI algorithm for identifying and quantifying several biomarkers of optical coherence tomography (OCT) in EMD, comparing the algorithm with human hand examination. Specifically, key biomarkers such as intraretinal fluid volume (IRF) and subretinal fluid volume (SRF) detection, outer limiting membrane (ELM), ellipsoidal zone (EZ) integrity and quantification of hyperreflective retinal foci (HRF) were considered. More than 100 eyes with EMD were included in the study. The accuracy of quantification of IRF, SRF, ELM and EZ by AI ranged between 94.7% and 95.7%, while the accuracy parameters ranged between 99.0% (OCT layer segmentation) and 100.0% (fovea centring). The intraclass correlation coefficient between clinical and automated HRF counting was excellent (0.97).
According to the results, there is therefore almost perfect agreement between the AI software and the human clinical evaluation for SRF volume and ELM and EZ integrity. Furthermore, the analysis comparing the number of HRFs assessed by AI and the clinical evaluation showed excellent reliability. In essence, the results of this study suggest that AI algorithms are reliable and reproducible tools to detect and quantify various OCT biomarkers in EMD, which are also very useful for a prognostic evaluation of treatment outcomes. In particular, artificial intelligence may facilitate the daily quantification of these biomarkers, as it has been shown to be as accurate and precise as clinical assessment, but less time-consuming.Â
Further studies will be needed to implement artificial intelligence software on a large scale, with real world data, to assess changes over time and the clinical relationship between changes in biomarkers and the course of the disease.
English 
Italian