Comparison of Transferred Deep Neural Networks in Ultrasonic Breast Masses Discrimination

Table 1

Features extracted in traditional model.

Feature category	Feature description
First-order features	FT1: Energy, FT2: Entropy, FT3: Minimum, FT4: 10th percentile, FT5: 90th percentile, FT6: Maximum, FT7: Mean, FT8: Median, FT9: Interquartile Range, FT10: Range, FT11: Mean Absolute Deviation (MAD), FT12: Robust Mean Absolute Deviation (rMAD), FT13: Root Mean Square (RMS), FT14: Standard Deviation, FT15: Skewness, FT16: Kurtosis, FT17: Variance, FT18: Uniformity
Texture features	Global texture features: FM1: Variance, FM2: Skewness, FM3: Kurtosis; Gray-Level Cooccurrence Matrix (GLCM) texture features: FM4: Energy, FM5: Contrast, FM6: Correlation, FM7: Homogeneity, FM8: Variance, FM9: Sum Average, FM10: Entropy, FM11: Dissimilarity, FM12: Autocorrelation
Morphological features	FB1: Circularity, FB2: Elongation, FB3: Compactness, FB4: Roughness, FB5: Orientation, FB6: Radial distance standard deviation, FB7: Maximum chord length, FB8: Second moment