Table 1.

Segmentation performance analysis by area normalized error. The first two rows represent the mean false positive and mean false negative values. The remaining rows report the distribution of the false positive and false negative values. For all the methods except the MST algorithm, the values reported are based on 84 mammograms. For the MST algorithm, the values reported are based on 82 mammograms (as the method failed to identify the pectoral muscle in two mammograms).

	Hough Transform	Gabor Wavelets	Adaptive Pyramid	Minimum Spanning Tree	Graph Cut Based Merging	Proposed Approach
${\text{FP}}_{\mathrm{m}}$	0.0198	0.0058	0.0371	0.0255	0.0064	0.0085
${\text{FN}}_{\mathrm{m}}$	0.2519	0.0577	0.0595	0.1168	0.0558	0.0488
$\text{FP}<0.05\hspace{0.17em}\&\hspace{0.17em}\text{FN}<0.05$	10	45	50	40	43	46
$\text{min}(\text{FP},\text{FN})<0.05\hspace{0.17em}\&\hspace{0.17em}0.05<\text{max}(\text{FP},\text{FN})<0.10$	0	0	18	20	19	25
$\text{min}(\text{FP},\text{FN})<0.05\hspace{0.17em}\&\hspace{0.17em}\text{max}(\text{FP},\text{FN})>0.10$	0	0	11	18	22	13
$0.05<\text{FP}<0.10\hspace{0.17em}\&\hspace{0.17em}0.05<\text{FN}<0.10$	8	22	0	0	0	0
$0.05<\text{min}(\text{FP},\text{FN})<0.10\hspace{0.17em}\&\hspace{0.17em}\text{max}(\text{FP},\text{FN})>0.10$	0	0	0	1	0	0
$\text{FP}>0.10\hspace{0.17em}\&\hspace{0.17em}\text{FN}>0.10$	66	17	5	3	0	0