Abstract: The digital mammary machine output of an original image is classified as two parts, background region and breast area. so as to help radiologist screening, and it is also an essential preprocessing step for computer aided diagnosis (CAD). This paper pro-posed a fast classification mask construction algorithm. First, initial breast segmentation with optimal threshold combined with an area classifier at low image resolution was performed, then Dijkstra searching method was used to retrieve the accurate breast profile at the o-riginal resolution, finally, a classification mask for the mammogram was constructed. Experiment results show this algorithm can extract breast profile region from raw mammogram data fast, robustly and accurately.