the and RG methods are given in table 2.

the tumor area is computed for each images. The results obtained from simulation are tabulated in table1. The relative error (RE) of each method for different tumor area is computed as:
?
A ? A’
?
RE (%) = ?
?
? 100
(17)

?
A
?
?
?
where A is the tumor area measured by different approaches
and A’ represents the tumor area as furnished by an expert radiologist. The comparison of relative errors using FCM, KM, MCWS, and RG methods are given in table 2. After
applying each algorithm one by one on all mammographic images, following statistics parameters are evaluated for further analysis 18:
(i) Sensitivity = Detected true positives / Real number of positives.
(ii) Number of false positive per image = Total number of false alarms / Total number of images.
Tumor area with relative error less than 50 percent is considered as detected ‘true positives’, otherwise as ‘false positives’. Therefore, in true sense, sensitivity presents the tumor detection rate for each method considering only those tumors whose segmented area matches at least 50 percent of the area furnished by the expert radiologist. All the segmented objects other than tumor are also considered as false alarms. Table 3 gives sensitivity, number of false positives per images and the relative error. Segmentation results are rated in one of the five possible labels: VG ? very good (with relative error range, 0-10%), G ? good (with relative error range, 11-20%), AVG ? average (with relative error range, 21-30%), BAVG ? below average (with relative error range, 31- 42%), B ? bad (with relative error range, >42%). From the table3, it is observed that, RG method yielded better results compared to other methods, with 72.20% of segmented masses rated very good (VG). Its score was one (B = 1) for bad cases. However, KM method achieved worst score with 4 bad cases, and 7 very good cases. Number of bad cases (B = 4), and very good cases (VG = 12), both were high in case of FCM method. MCWS got the average score with 3 bad cases, and 10 very good cases. From the above discussion, it can be said that RG method segmented the tumors more accurately than the others with the least deviation from the actual one (outlined by an expert radiologist) in maximum number of cases (VG = 13). After RG method, FCM method achieved the higher segmentation accuracy of the extracted tumors with 12 very good cases. . The FCM achieved the sensitivity of 77.80% with 2.50 false positives per image, KM with sensitivity of 77.80% at the rate of 3.92 per image, MCWS with sensitivity of 83.30% at the rate of 3.50 per image, and with sensitivity of 94.44% at the rate of 1.12 per image by RG method. Thus, RG method seems to provide good segmentation results with 94.44% sensitivity at the rate of 1.12 FP/image, and with maximum number of cases segmented with greater accuracy than any other method presented here.

x

Hi!
I'm Marcella!

Would you like to get a custom essay? How about receiving a customized one?

Check it out