We aimed to develop and test a deep-learning system to perform image quality and diabetic macular ischemia (DMI) assessment on optical coherence tomography angiography (OCTA) images.
This study included 7,194 OCTA images with diabetes mellitus for training and primary validation and 960 images from three independent data sets for external testing. A trinary classification for image quality assessment and the presence or absence of DMI for DMI assessment were labeled on all OCTA images. Two DenseNet-161 models were built for both tasks for OCTA images of superficial and deep capillary plexuses, respectively. External testing was performed on three unseen data sets in which one data set using the same model of OCTA device as of the primary data set and two data sets using another brand of OCTA device. We assessed the performance by using the area under the receiver operating characteristic curves with sensitivities, specificities, and accuracies and the area under the precision-recall curves with precision.
For the image quality assessment, analyses for gradability and measurability assessment were performed. Our deep-learning system achieved the area under the receiver operating characteristic curves >0.948 and area under the precision-recall curves >0.866 for the gradability assessment, area under the receiver operating characteristic curves >0.960 and area under the precision-recall curves >0.822 for the measurability assessment, and area under the receiver operating characteristic curves >0.939 and area under the precision-recall curves >0.899 for the DMI assessment across three external validation data sets. Grad-CAM demonstrated the capability of our deep-learning system paying attention to regions related to DMI identification.
Our proposed multitask deep-learning system might facilitate the development of a simplified assessment of DMI on OCTA images among individuals with diabetes mellitus at high risk for visual loss.