[Study of ultrasonographic image textures with second-order statistical methods].

Abstract

BACKGROUND: We report the sonographic image texture of the neonatal heart in different stages of development by calculating numerical parameters extracted from the gray scale cooccurrence matrix. To show pixel values differences and enhance texture structure, images were equalized, and then the gray level range was reduced to 16 to allow for sufficiently high occupancy frequency of the cooccurrence matrix. MATERIALS AND METHODS: We measured B-mode US images, all acquired with the same unit, using the same setup (gain, frequency, constant TGC). For each case measurements were made on 3 diastolic images by selecting a 32 x 64 pixel on the interventricular septum ROI. This procedure was applied to 3 different examinations made at birth, at 3 days, and after a month of life. For each 8 bit image we obtained a coocurrence matrix by sampling adjacent pixels at 0 degree, from which we calculated entropy; then the images were equalized and converted to 4 bit format; on these processed images we calculated 4 gray scale coocurrence matrix samplings adjacent pixels at angles of 0 degree, 45 degrees, 90 degrees, and 180 degrees. Parameters obtained from different angles cooccurrence matrix were averaged to get a numerical parameter independent of sampling angles. RESULTS: The one-way analysis of variance and Student's "t"-test were performed on the parameters obtained from 3 different examinations and the differences among different developmental stages are 95% significant. No significance was obtained comparing images at birth with those at 3 days. CONCLUSIONS: Differences are so little significant that they may be due to different factors affecting image texture and the variability introduced by manual ROI positioning; therefore no definitive conclusions can be drawn as to considering this kind of analysis capable of discriminating different stages of myocardial development. To apply this analysis to routine US examinations, different correction factors for range, azimuth, and focal dependence of texture must be considered.