Ultrasonic image parameters for differentiating intraocular melanoma

Abstract

In this paper in -vivo differentiation of t > p s of intraocular melmoma using texture parameters is reported. Differentiation of various cliIssCs of malignant melanoma through imaging is of special clinical interest for patient treatment and follon up. Texture parameters based on first and second order statistics and run length statistics o f clinical ocular B-Scan imilges are utilized to obtain the quiintitzitive information using image processing algorithms. The results are compared for tumors of different cases: spindle and mixed cell melanomas. Ultrasonic B-Scan images were obtained using a commercial ultrasound scanner. Results: show that the spindle cell melanoma has greater reflectivity and i t is more homogeneous than the mixed cell melanoma and i t is possible to differentiate the spindle and mixed cell melanoma using some of the texture par:irneters. keywords: Imaging, Melanoma, Texture, U1 trasound INTROD C'CTIOMalignant Melanoma is an important intraocular tumour to be diagnosed early and is the commonest intraocular malignancy in adults. Intraocular melanoma of the choroid located posterior to the retina and anterior to the sclera have been found to lead to retinal detachment and blindness. It is a brown, elevated convex iiiass and is diagnosed b! . ullrasound imaging by its location and special shape of 'niusliroom'.Diffcrentiation of malignant tjpe of intraocular meianoma is of special clinical interest for patient-treatment and patient-follow-up. The four niost important determinants of prognosis in the tumour are the size, the cell type, mitotic activity and presence or absence of extrascleral extension [I].Of these diagnostic predictors, Ultrasound can guide tlie clinician in judging the tumour size and presence or absence of estrascleral extension. Tlie segregation of sub h p e s of melanoma is not possible by routine ultrasonic imaging. It is important to know the cell type since cell t5pe is a strong indicator of risk and no dcatlis occurred when epitlicloid cell were not present 131. For this purpose the standard ultrasound equipment is not suficient and methods other than the visual assessment have to be used.To diffcrentiatc the tunior from the ultrasonic scan image ti:xture parameters are utilised. Texture is concerned with the spatial (statistical) distribution of gray tones [j]. Statistical pattern recognition procedures allow a quantitative description of ultrasonic B-scan image texture. The texture parameters i.e.. first and second ordcr statistical parameters and run length-statistical parameters can be uscd to describe quantitatively tlie microstructure and macrostructure of the ultrasonic image [?I. Tlie first-ordcr texture is sufficiently characterised by the gray level histogram, \vliicli displays the occurrence frcquency of all gray Icvcls in the image cluster. The second-order statistics describe not only the occurrence frequency of gray le\tls, but also uses spatial interdependencies between the image clcments. Due to a lack of spatial information tlie parameters taken from t...