Deep learning for OCT pathology recognition

We are pleased to report that our abstract, “Deep convolutional neural networks for automated OCT pathology recognition” was accepted for presentation at the 2017 Annual Meeting of the Association for Research in Vision and Ophthalmology (ARVO).  The research was performed in collaboration with Dr. Robert T. Chang of Stanford University’s Byers Eye Institute.

Deep convolutional neural networks or, as they have become more popularly known, “Deep Learning”, have been a breakthrough in the field of image recognition and classification. Medical images, however, are different from natural images and have different relevant features sets requiring more sophisticated forms of preprocessing. We found that, using automated retinal layer segmentation-based preprocessing together with a deep convolutional neural network, we could achieve excellent results in the task of classifying a given optical coherence tomography (OCT) B-scan as pathological or normal. One of the advantages of this technique is that we don’t have to specify the disease. Future applications of this work could result in a fully-automated front-line OCT screening tool which could refer patients to ophthalmologists only when necessary. Look for this technology to be integrated into Orion^TM at Voxeleron’s booth at ARVO 2017.