Structural image texture and early detection of breast cancer

Structural texture measures are used to address three aspects of early detection of breast cancer in screening mammograms: detection of microcalcification, detection and classification of clustered microcalcification as benign or malignant, and the detection of invasive lobular carcinoma. The use of structural texture features replaces the task of initial detection of complex and poorly modelled image structures such as masses or clustered microcalcifications by initial detection of primitive image structures. Receiver operating characteristic (ROC) analysis yields high performance scores for detection of microcalcification ( ) and for classification of clusters as benign or malignant ( ). In the case of invasive lobular carcinoma, cancer was detected in half of the images with no false positive detections. In these cases, no evidence of cancer could be found by visual inspection of the mammogram, thus demonstrating the potential of structural texture measures to encompass diagnostically useful information not accessible to the human expert observer.