Evaluation parameters and technique for classification of intraocular tumours

Classification and diagnosis of intraocular tumours is one of the actual problems in ophthalmology. Melanoma, metastasis and no malignant tumour are frequently observed intraocular tumours. Simply no malignant tumour in the text refers as “no tumour”. Melanoma is melanocytus neoplastic tumour, which may induce the secondary tumours (metastasis) to lungs, liver and bones. Melanoma commonly starts to grow in the eye posterior urea. Melanoma is the direct reason of death for each third patient. Therefore early diagnosis of this pathology is crucial for proper treatment and prognosis. Ultrasound echographic technique is widely used for visualization of intraocular tumours. Non invasiveness, informativity and user-friendliness are the main features of ultrasound technique. The brightness of ultrasound image element (pixel) depends on the distribution of acoustic impedance or tissue density. The possibility of ultrasound technique applying for classification of various tumours is based on the direct relationship between tumour tissue density distribution and its morphology. Set of publications is made on echographic characterization of intraocular tumours [1, 2, 3]. Ordinary method for analysis of echographic images is based on search and evaluation of specific parameters, statistical analysis and methods of image texture classification. A 15-year retrospective analysis of tumours classification is presented in [1]. Main echographic parameters are determined in [3]. Application of data mining technique for medical decision support is discussed in [4]. Fundamental technique for parametrisation of ultrasonic images is presented in [5]. The essence is to use the point spread function and inverse filtering technique for restoration of ultrasonic images. Further investigations are needed to show whether the use of complicated mathematical tools allows to reduce the image noises and gives a more useful information. Analysis of publications shows that the possibilities of ordinary methods for tumour classification are not exhaused. It is desirable to search for new specific parameters, based on the information about the mechanism of image element genesis, to determine interrelationship and to increase the reliability of classification. The aim of the present paper is to develop technique, to find optimal values of specified ultrasonic images parameters and investigate the system See5.0(C5.0) for classification of intraocular tumours.