Application of Artificial Neural Networks for Direct Detection of Microcalcification Clusters in Digital Mammograms

Abstract

Computer-aided diagnosis (CAD) schemes for the detection of microcalcification clusters (MCCs) come in two types: indirect and direct. Indirect detection of MCCs detect individual microcalcifications first, which are then used to detect clusters. Direct detection detects clusters in a unique step, without any previous detection of individual microcalcifications. Nearly all the existing literature describes indirect detection. In this study, we investigated a direct detection scheme. We divided digital mammograms into regions of interest (ROis) and computed a set of parameters on each ROI. We discriminated parameters through an artificial neural network (ANN) that gave the presence or absence of an MCC in the examined ROI. Final images with suspicious ROis containing MCCs were shown to radiologists. Results appeared to be interesting enough to compete with indirect detections. Extra studies could prove direct detection to be a better approach as compared to indirect detection CAD schemes.