Automated classification of ocular ultrasonography using CNN

Inhalt:
Ocular ultrasonography interpretation is performed mainly by ophthalmologists, which will take up a lot of time in diagnosis. CNN has not been used to classify ocular ultrasonography. We aim to use deep learning to classify the ocular ultrasonography into seven categories. For this purpose, we introduce an ocular ultrasonography (OUG) dataset which is still extended. The dataset is composed of groundtruth annotation as well as several categories diagnosed by five professional ophthalmologists. We test some models and apply data augmentation strategy to improve the performance. The open-source deep learning framework PyTorch is used to accelerate the deep learning process. After several tests with an iteration of 200 training steps, we choose GoogleNet as our classification model. The validation accuracy of classification is 78.7%.