Artificial intelligence and glaucoma structure-function map: a systematic review

With advancing insights into the structure-function relationship in glaucoma, developing visual field (VF) predictive models based on optical coherence tomography (OCT)-detected structural changes have emerged as a significant research hotspot. Artificial Intelligence (AI) technologies have been widely applied in this area, demonstrating significant potential. A comprehensive search was conducted using databases such as PubMed, Scopus, Web of Science, and Embase with key search terms including “artificial intelligence,” “machine learning,” “deep learning,” “optical coherence tomography,” “visual field,” “structure-function map,” and “glaucoma”. Following rigorous and meticulous inclusion, exclusion, and review processes, a total of 36 studies were ultimately selected for analysis. Several studies reported favorable performance for predicting global VF indices, pointwise VF sensitivity, or longitudinal VF progression, although performance varied across tasks, datasets, OCT devices, VF patterns, validation designs, and performance metrics. Various visualization techniques, including heatmaps, simulation of structure-function maps, and occlusion maps, have been employed to interpret the predictions of these models. The generated structure-function maps are consistent with the current clinical understanding of the relationship between glaucoma structure and function. However, current AI models generally suffer from decreased performance in the advanced stage glaucoma and are limited by the narrow scanning range of OCT. Current glaucoma structure-function AI prediction models have been applied to pointwise VF prediction. However, to further deploy these models in glaucoma clinical settings, it is necessary to construct models that can predict pattern deviation probability maps based on wide-field volumetric OCT scans.