Current CAD systems always demand better performance, both in terms of the best sensitivity-specificity tradeoff and of the processing time. In order to decrease the false positive rate and to increase the time responsiveness of our CAD system, we present a powerful algorithm that performs an intra-breast segmentation. Starting from a digital mammogram and by using a combination of different techniques, the proposed algorithm produces a binary map showing the locations of the breast with higher probability of containing a suspect mass. Experimental trials performed on 317 mammograms show that our algorithm is able to reduce the intra-breast searching area down to 6% of the entire breast area, while maintaining almost the same negative rate. Typical dimension of images is about 2800×2000 pixels at 13 bits of gray level resolution. The set of images was gathered from FFDM units coming from two different hospitals.
A fast algorithm for intra-breast segmentation of digital mammograms for CAD systems / R. Campanini; E. Angelini; E. Iampieri; N. Lanconelli; M. Masotti; M. Roffilli; O. Schiaratura; M. Zanoni. - ELETTRONICO. - (2004). (Intervento presentato al convegno International Workshop on Digital Mammography 2004 tenutosi a Durahm (USA) nel 18-21 giugno 2004).
A fast algorithm for intra-breast segmentation of digital mammograms for CAD systems
CAMPANINI, RENATO;LANCONELLI, NICO;MASOTTI, MATTEO;ROFFILLI, MATTEO;
2004
Abstract
Current CAD systems always demand better performance, both in terms of the best sensitivity-specificity tradeoff and of the processing time. In order to decrease the false positive rate and to increase the time responsiveness of our CAD system, we present a powerful algorithm that performs an intra-breast segmentation. Starting from a digital mammogram and by using a combination of different techniques, the proposed algorithm produces a binary map showing the locations of the breast with higher probability of containing a suspect mass. Experimental trials performed on 317 mammograms show that our algorithm is able to reduce the intra-breast searching area down to 6% of the entire breast area, while maintaining almost the same negative rate. Typical dimension of images is about 2800×2000 pixels at 13 bits of gray level resolution. The set of images was gathered from FFDM units coming from two different hospitals.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.