Retinal Feature Quantization Method Based on a Reference Model
TONG He-jun,FU Dong-mei
DOI: https://doi.org/10.13374/j.issn2095-9389.2019.09.015
2019-01-01
Chinese Journal of Engineering
Abstract:Optical coherence tomography (OCT) plays an important role in the diagnosis of ocular fundus diseases. Retinal OCT images contain a large amount of useful information for the diagnosis of ocular fundus diseases and are often used to detect small lesions of the fundus. At present, many medical researchers have used OCT to determine the statistical characteristics of the retina to analyze various fundus diseases. When interpreting the OCT images, ophthalmologists will focus on the location of the lesions in the images and the characteristic morphology conducive to abnormal judgment and compare the histological structure of specific objects in the images with the known normal morphology. In the comparison process, the ophthalmologist will conduct a variety of quantitative analyses of OCT retinal images and determine the severity of the abnormalities and the location of the lesions. Finally, on the basis of the differences between the morphologies and types of diseases, the diagnostic decision is obtained. However, at present, OCT instruments generally only provide the thickness, area, and other commonly used characteristic data, and these data are often inadequate to determine the disease. Computer graphics processing technology has been applied to the auxiliary analysis of OCT images. However, this kind of research often confines the object of study to several specific fundus diseases and makes targeted selection of quantitative features. In the actual diagnosis process, it is difficult to confine the retinal images to some known abnormal cases because of the complexity of the situation. In this study, a retinal feature quantization method based on a reference model was proposed, and a series of quantifiable features suitable for computer judgment and analysis of retinal state were proposed. On the basis of the segmentation and extraction of the internal limiting membrane (ILM), junctions of the inner and outer segments of photoreceptors (ISOS) and Bruch's membrane (BM) in normal OCT images, a reference model of normal retina was constructed by the statistical method. Combining the reference model with the retinal thickness, smoothness, and continuity, the thickness characteristics, thickness ratio characteristics, gradient characteristics, curvature, standard deviation, and correlation coefficient characteristics of different regions of the retina were calculated. On the basis of the reference model of normal OCT images, the quantitative values of retinal thickness and morphological characteristics were obtained. By analyzing and comparing the characteristic value differences between abnormal OCT images and reference model, the location and severity of abnormal morphology caused by lesions could be characterized in the abnormal OCT images. The experimental results show that the normal retinal feature information obtained by the reference model can provide a numerical reference for ophthalmologists. At the same time, the characteristic values obtained by quantizing the abnormal OCT images can show the abnormal morphology, which provides a basis for subsequent abnormal judgment.