This computerized system accurately guides laser shots to diseased retinal areas.
Diabetic retinopathy resulting from long-term diabetes mellitus is one of the common diseases that leads to choroidal neovascularization (CNV), a leading cause of blindness. Among the currently available treatment methods, a laser can be used to photocoagulate the diseased areas. Several thousand laser shots are usually required during such treatment. Special care must be taken to avoid hitting the blood vessel tree, the macula, the optic disk, and the region among them. For a single eye, this procedure requires up to several hours that are usually divided over many treatment sessions. Consequently, the development of an accurate laser treatment guidance system to treat the whole retina in one session would improve the effectiveness of such procedures.
A new computerized technique for laser treatment planning is proposed. It performs accurate segmentation on a reference frame. The locations of the laser shot on this reference image are then recomputed for the sequence of images acquired during the treatment using fast registration. This enables fast tracking of retinal structures and ensures proper administration of the treatment in case of eye movement. This new system has the potential to improve the accuracy of the procedure, thus lowering its failure rate while maintaining the same treatment time.
A major part of this system is the tracking of the retina using a computer-controlled system that captures and processes retinal images. Novel image processing routines are applied to the retinal images to determine the correct positions for laser shots. An accurate segmentation procedure is applied to the first retinal image to extract the sensitive areas in the retina. Subsequently, this segmentation is continuously updated using a fast registration procedure to obtain the new positions in case of eye movement at real-time rates. These positions can be fed to a beam-steering apparatus that precisely directs the laser beam accordingly.
The retinal images are acquired using a fundus camera. Contour detection is performed using deformable models. This technique is an iterative process in which one initiates an active contour model close to the contour of interest. The contour model is deformed using one of many algorithms for searching the neighboring points of each snaxel. All the points having a value above a certain threshold are extracted as new points on the contour. This not only deforms the contour but also makes it grow.
After extracting all sensitive objects that must be avoided during laser treatment, a binary image was composed containing these objects. This image was dilated by a square structuring element of dimensions 7×7 to maintain a safety margin around these sensitive areas. These locations are stored in the patient database, and get updated with every successive image frame.
A fast registration technique is proposed to automatically track the motion of segmented areas within the subsequent images in the sequence. In this technique, points satisfying certain developed significance conditions are chosen as landmarks in the reference frame. Using the extracted set of corresponding points in the subsequent image frames, one can accurately align image frames to compensate for eye movements and saccades.
This work was done by Nahed H. Solouma, Abou-Bakr M. Youssef, Yehia A. Badr, and Yasser M. Kadah of Biomedical Engineering Department and Laser Institute at Cairo University for the Army Research Laboratory. ARL-0064