Image processing and hierarchical temporal memories for automated retina analysis

Due to the projected increase in the type 2 diabetes epidemic, there is a critical need for widely available and inexpensive screening for diabetic retinopathy, a preventable secondary disease caused by diabetes that can lead to decreased visual function and even blindness. Currently this type of testing can only be performed manually by ophthalmologists, but a telemedicine network with retina cameras and automated quality control, physiological feature location, and lesion / anomaly detection is a more cost effective method of providing broadbased screening. In this paper we report on the method of using Hierarchical Temporal Memories (HTMs), a new type of machine learning technology based on the function of the human neocortex, to locate optic nerves as an alternative method for physiological feature location as a part of the larger telemedicine network scheme. We compare the results from the HTM network on a data set collected from a Memphis, TN clinic to the results from more conventional machine vision techniques. We show that while HTM technology as it is used with this procedure is not as accurate as traditional image analysis and processing methods, it is still reasonably effective and is a promising new technology for machine vision applications such as the diabetic retinopathy telemedicine network.