Towards automated mammograph image analysis

Abstract

Two alternative practices are commonly followed when detecting and/or describing breast cancer tumors on mammography images. Medical radiologists normally describe the tumor in words, making reference to its mass, shape and margins. Meanwhile, pattern recognition specialists have their own methodologies. Since there are significant gaps between two approaches, it has proven to be very difficult for those following the pattern recognition route to directly adapt parameters of mass, shape and margins for the automated recognition of different cancers. This paper describes a joint R&D project of Yunnan University & Yunnan First People's Hospital. A meta-shape tool and conjugate meta-feature clustering technology have been developed. These represent initial steps in the descriptions of mass, shape and margins on the road towards possible automated mammograph image analysis. In this model, ten meta-shape feature clusters are used to provide a systematic means of representing different cancerous symptoms. To indicate potential applications, a group of selected results are outlined to illustrate possible linkages between the two approaches.