Table 6 The results applied using the SqueezeNet pre-trained CNN and ADAM optimizer per class.

From: Adapting the pre-trained convolutional neural networks to improve the anomaly detection and classification in mammographic images

CNN

Class

Performance of the Classifier

Accuracy (%)

Sensitivity

Specificity

Precision

AUC

Before pre-processing

Benign

63.9

0.37

0.69

0.41

0.45

Malignant

60.2

0.35

0.65

0.38

0.46

Normal

64.2

0.34

0.68

0.37

0.44

Average

62.76

0.353

0.673

0.386

0.45

After pre-processing

Benign

97.1

0.89

0.99

0.95

0.998

(SM)

Malignant

98.2

0.949

0.979

0.84

0.998

Normal

95.9

0.97

0.94

0.95

0.995

Average

97.06

0.936

0.969

0.91

0.997

After pre-processing

Benign

99.5

1.0

0.988

0.93

0.998

(MSVM)

Malignant

98.9

0.987

0.99

0.95

0.999

Normal

99.2

0.978

0.995

1.0

0.998

Average

99.2

0.988

0.991

0.96

0.998

After pre-processing

Benign

99.2

0.99

0.99

0.94

0.991

(RF)

Malignant

99

0.979

0.987

0.947

0.998

Normal

98.4

0.98

0.992

0.99

0.998

Average

98.86

0.983

0.989

0.959

0.995

Back to article page