Improving Minimum Cross-Entropy Thresholding for Segmentation of Infected Foregrounds in Medical Images Based on Mean Filters Approaches
Table 5
Overall comparison using average evaluation metrics for the proposed method using the alpha trim approach and two related methods. Row A refers to an unsupervised evaluation, and row B refers to the supervised evaluation.
Original MCET-Gaussian distribution
Dermoscopic skin lesion
MRI brain tumor
MRI Alzheimer’s
A
0.8703
0.8406
0.8841
B
0.8923
0.8210
—
MCET-lognormal distribution
Dermoscopic skin lesion
MRI brain tumor
MRI Alzheimer’s
A
0.8802
0.8693
0.8874
B
0.8817
0.8807
—
Overall increase rate of accuracy over Gaussian = 2.1%
Proposed MCET-alpha trim d/2 = 55
Dermoscopic skin lesion
MRI brain tumor
MRI Alzheimer’s
A
0.8917
0.8789
0.8968
B
0.9208
0.8839
—
Overall increase rate of accuracy over Gaussian = 3.7%
Overall increase rate of accuracy over lognormal = 1.6%
Proposed MCET-alpha trim d/2 = 50
Dermoscopic skin lesion
MRI brain tumor
MRI Alzheimer’s
A
0.9102
0.8779
0.90426
B
0.8876
0.8860
—
Overall increase rate of accuracy over Gaussian = 3.65%
Overall increase rate of accuracy over lognormal = 1.5%
Proposed MCET-alpha trim d/2 = 30
Dermoscopic skin lesion
MRI brain tumor
MRI Alzheimer’s
A
0.8739
0.8942
0.8928
B
0.9059
0.9165
—
Overall increase rate of accuracy over Gaussian = 4.0%
Overall increase rate of accuracy over lognormal = 1.9%
