Classification of optic disc pathologies in NEAR infrared reflectance images with DEEP learning

Objectives: The aim of this study is to develop an appropriate algorithm with artificial intelligence (AI) deep learning (DL) method by using near infrared reflectance (NIR) images of optic disc edema and pseudopapilledema and to evaluate the effectiveness of the developed model. Materials and Methods: NIR images of patients who underwent detailed ophthalmological examinations and were found to have optic disc edema or pseudopapilledema were examined. In the optic disc edema group, a total of 158 images of 24 patients with papilledema, 16 non-arteritic ischemic optic neuropathy (NAION), 12 central retinal vein occlusion (CRVO), 3 demyelinating optic neuropathy and 5 diabetic papillopathy patients were used. 346 pseudopapilledema images were obtained from the images of 70 patients with optic disc drusen and 62 patients with peripapillary hyperreflective ovoid mass-like structures (PHOMS). Also NIR images of 336 eyes of 168 healthy individuals without any optic disc pathology were included in the study. Infrared images were divided into 2 groups for training and testing of the model. 85% (714 images) of the images were used for training the model and 15% (126 images) for testing the trained model. A DL model was constructed using the convolutional neural networks (CNN) algorithm to classify optic disc images. Basically, the data was introduced to the model, allowing it to train itself and tested with the images reserved for testing. Results: The developed model was tested with 24 optic disc edema, 52 pseudopapilledema and 50 normal optic disc images not used in training. Sensitivity, specificity and accuracy of the model were calculated in detecting optic disc edema, pseudopapilledema and normal optic discs without optic disc edema or pseudopapilledema. Receiver Operating Characteristic (ROC) curve and area under the curve (AUC) values were also analyzed. Respectively; sensitivities were 100%, 98%, and 96% specificity, 99%, 97%, 100%, and accuracy rates were 99%, 98%, and 98%. In addition, the auc value of the groups was 0.995 (95% Confidence interval [CI]: 0.98-1); 0.983 (95% CI: 0.96-1); 0.973 (95% CI: 0.94-1). Conclusion: The results show that the DL model developed for the diagnosis of optic disc pathologies can be used with high efficiency in the detection of optic disc edema and pseudopapilledema. Optic disc pathologies can be diagnosed by integrating similar AI models into OCT devices capable of NIR imaging. In order to develop the model, it needs to be supported by other imaging methods and more images are needed. Keywords: Artificial intelligence, Deep learning, Optic disc edema, Pseudopapilledema, Near infrared reflectance