[Texture analysis: a new method for evaluating ultrasound imaged lesions of the breast].

Abstract

Echogenicity and echostructure in sonomammography are essential criteria to determine the malignancy of a breast tumor. Looking for a correlation between echostructure of an ultrasound image and its histopathology a texture analysis was performed. Ultrasonic examinations were carried out by means of a 7.5-MHz linear scanner under standardized conditions. 71 mammasonographic findings were documented by video tape, each tumor in 2 dimensions. In a second step 'the region of interest', that means the whole tumor area, was marked and statistically analyzed by a dedicated computer system. The values of 23 first- and second-order statistical texture parameters were compared with the histopathology of the tumor. They include grey level histogram, Fourier analysis and cooccurrence matrix. The most important results were obtained in the histogram analysis. An obvious distinction could be worked out concerning carcinomas, adenomas, mastopathic tumors, and fat necrosis. Except the difference between carcinomas and adenomas this distinction was statistically significant. Fourier analysis and coocurrence matrix significantly separate fibrocystic lesions, fat necrosis, scars, and cystic structures from malignant tumors. Furthermore the maximum value method was used to compare the different parameters with regard to their capability of discriminating benign and malignant tumors. The extreme values of 100% were mostly observed in the case of seromas and scars. A significant difference was found by histological grading of carcinomas, too. Highly differenciated carcinomas show capability of results comparable to those of adenomas. By means of a computer texture analysis program it is possible to find a correlation between echostructure and histopathology of breast tumors.(ABSTRACT TRUNCATED AT 250 WORDS)