Texture analysis of ultrasound B-scans to aid diagnosis of cancerous lesions in the breast

Ultrasound (US) B-scans in conjunction with mammography reduce the false positive rate intrinsic to mammographic diagnosis by separating cysts from solid tumors. A description is given of the Phase I study whose intent is to decrease further the false positive rate through quantitative analysis of the solid tumors as they appear in US. Two sets of texture parameters were used: run length and Markovian. A two-dimensional combination of one parameter of each set was chosen as the best classifying criteria. A preliminary study of 118 tumors, of which 35 were malignant, shows an improvement of 40% in the false positive rate of the cases whose US was read. It also shows that the texture analysis system (TAS) is far superior to a human professional reader working with the same data.