Spicule descriptors designed in complex wavelet domain to recognize breast cancer symptoms

The total number of words in the manuscript, including entire text from title page up to figure legends: 7085 The number of words of the abstract: 200 Abstract The subject of presented paper is effective recognition of radiating spicules on digital mammograms. The presence of the spicules is a dominant symptom of neoplastic breast lesionscalled architectural distortions. The originality of the proposed method is the extraction of effective descriptors concentrated with the local directional activity of mammographic texture. Applied methodology was based on the analysis and constructively modeling the conditioning of spicules distribution in complex wavelet domain. It is because of potentially acquisition-invariant energy compaction across directionally spread, especially for piecewise linear structures represented across scales, directions and locations of normalized domain. Additionally, the use of non-directional properties of mammographic findings completes essential conditioning of abnormal spicule appearance. Adaptive analysis of breast tissue distribution in complex wavelet domain allows separating specific local centers of increased multidirectional texture activity among slightly varied but the dominant tendency of tissue directional distribution. Optimized and empirically verified numerical descriptors of local tissue spiculation were fundamental aspect of the proposed effective method to recognize radiating cancer symptoms. Experimental study with a test set of 280 regions of interests, containing normal and abnormal breast tissue of clinically confirmed ADs, have revealed the recognition sensitivity and specificity close to nearly 77% and over 73.5%, respectively.