New texture shape feature coding-based computer aided diagnostic methods for classification of masses on mammograms

This paper presents new texture shape feature coding (TSFC)-based computer aided diagnostic (CAD) classification methods for classification of masses on mammograms. It introduces a new concept of 1.5-order 3-neighbor 3 × 3 connectivity to extract texture shape features that can describe multiples of 22.5°. In order to effectively utilize these shape features, two new methods of implementing TFSC are further proposed to convert these features to texture feature numbers (TFNs), TFNs in quaternary (TFNq) which expresses a TFN in quaternary expansion and TFNs in product (TFN × ) that represents a TFN in terms of a product. Both TFNq and TFN × can then produce texture shape histograms in the same way that a gray-level histogram is generated for an image. Such a texture shape histogram is further used to generate various shape features of masses on mammograms for classification. In order to demonstrate the promise of our TSFC-based CAD methods, the MiniMammographic Database provided by the Mammographic Image Analysis Society (MIAS) is used for experiments.