Automatic recognition of malignant lesions in ultrasound images by artificial neural networks

A method for the automatic classification of lesions in ultrasound images by artificial neural nets is presented. The parameters used for training of the network are texture related indicators and shape related indicators. Three lesions have been considered: cysts, fibroadenomas and carcinomas. Solid lesions have been separated from cysts in the first step and carcinomas have been separated from fibroadenomas in a second step. A satisfactory classification between cysts and solid lesions can be achieved using texture parameters only, whereas shape parameters appear to be the most significant ones when classifying between carcinomas and fibroadenomas.