I received this via an email update earlier on (from A*Star) and thought I’ll share it.
The human eye has a blind spot, a region where optic nerves meet and therefore has no photoreceptors for detecting and perceiving light. This blind spot, also known as the optic disc, plays a crucial part in the eye’s physiology and the diagnosis of eye diseases. However, optic disc detection and segmentation from retinal images can become challenging due to various ocular pathologies that could degrade the image quality severely.
Shijian Lu at the A*STAR Institute for Infocomm Research and co-workers have now provided a solution to this long-standing problem by developing a computer algorithm that is able to detect the optic disc from retinal images with unprecedented precision and accuracy1.
The variations in optic disc appearance for different eyes have made it difficult for computer algorithms to pinpoint disc centre and boundary with sufficient accuracy for medical diagnostics. Often, diseases or other features in the eye such as blood vessels make assignments difficult. The basis on which algorithms identify the optic disc is usually through its brighter appearance compared to surrounding areas. Through such an analysis, a region can be identified in which the optic disc is most likely to be.
The algorithm developed by Shijian Lu now takes the information on the probable locations for the disc and refines it by taking a step further — assuming that the optic disc is usually round. The circular transformation method developed by Lu looks for maximum variations in brightness along radial lines spreading out from the region of the probable location of the optic disc. By passing through several filters, the researchers could identify the disc boundary, and consequently the disc center.
In tests on standardized retina photographs, the algorithm was able to identify the optic disc with 98.8% detection accuracy. The placement error of the disc center was only six pixels. Moreover, the sampling speed of the photos was only five seconds. This can be enhanced even further by at least a factor of ten as the software was written on a non-optimized software package.
Such accuracy and sub-second speeds make this method promising for clinical use. “This is a breakthrough for automatic computer aided diagnosis of ocular diseases, because few state-of-the-art techniques can handle the optic disc segmentation for severely degraded pathological retinal images,” says Lu. Clinical trials under more difficult circumstances than the standardized photographs will follow. If successful, this new method could greatly improve the detection of eye diseases.
Just an FYI, I actually reviewed the system used to acquire and distribute the images (it was part of a tele-ocular project) and the only missing piece (back then) was the computer aided diagnosis part of things and viola, seems like they are close to sorting this out.