Automatic Segmentation of Periapical Radiograph Using Color Histogram and Machine Learning for Osteoporosis Detection
Table 3
Training performance of K-means and Fuzzy C-means classification.
Segmentation
ML classifier method
Performance
K
8
10
12
15
K-means
Decision tree
Accuracy
86.67
88.33
80
75
82.50
81.67
Specificity
83.33
90
76.67
73.33
Sensitivity
90
86.67
83.33
76.67
Naive Bayes
Accuracy
75
83.33
76.67
73.33
77.08
Specificity
73.33
80
80
66.67
Sensitivity
76.67
86.67
73.33
80
Multilayer perceptron
Accuracy
88.33
91.67
80
81.67
85.42
Specificity
86.67
90
76.67
83.33
Sensitivity
90
93.33
83.33
80
Fuzzy C-means
Decision tree
Accuracy
78.33
75
63.33
71.67
72.08
75.28
Specificity
83.33
70
70
76.67
Sensitivity
73.33
80
56.67
66.67
Naive Bayes
Accuracy
80
70
73.33
73.33
74.17
Specificity
80
73.33
63.33
70
Sensitivity
80
66.67
83.33
76.67
Multilayer perceptron
Accuracy
81.67
68.33
83.33
85
79.58
Specificity
83.33
70
80
83.33
Sensitivity
80
66.67
86.67
86.67
The average performance describes the average of all performance characteristics (accuracy, specificity, or sensitivity) of an ML classifier. The global performance is derived from the average of all parameters (accuracy, specificity, and sensitivity) of segmentation methods utilizing all ML classifiers.
