MEASUREMENT OF CHOROIDAL VASCULARITY INDEX AND CHOROIDAL THICKNESS USING DEEP LEARNING-BASED METHODS WITH SWEPT-SOURCE OPTICAL COHERENCE TOMOGRAPHY

PURPOSE: In this study, we aimed to fully automate the measurement of choroidal thickness and choroidal vascularity index in images obtained using swept source optical coherence tomography with the aid of deep learning-based artificial intelligence. METHODS: Seven hundred and ninety-one swept source optical coherence tomography images obtained from 652 patients were included in the study. Optical coherence tomography images of healthy people or patient with ocular disease aged 18 years and older were included in the study. Of the data set created by labeling 791 images, 474 (59%) were used for model training, 237 (29%) images were used for validation, and 80 (10%) images were used for testing model. An artificial intelligence model was created using the convolutional neural network algorithm to detect the choroidal region and calculate the choroidal thickness. Then, choroidal vascularity index was calculated with the threshold method. Labeled images were introduced to the model, allowing the model to train itself and then testing the model with the images reserved for testing. RESULTS: The model was tested with 80 images. Accuracy, mean absolute error, intersection over union, and area under curve metrics were used to evaluate the model. While the accuracy value of the model in calculating the choroidal region thickness was found to be 0.96955, the mean absolute error was found to be 0.0286, intersection over union was 0.9249, and area under curve was 0.9941. In the evaluation made between two experienced ophthalmologists and the model for choroidal thickness, the intraclass correlation coefficient value was found to be 0.987 (0.979-0.993). CONCLUSION: The results obtained showed that the calculation of choroidal thickness and choroidal vascularity index can be automated with the help of artificial intelligence and be highly effective.