A machine learning approach for differentiating malignant from benign enhancing foci on breast MRI

European Radiology Experimental

Table 3 Features extracted using the TWIST (training with input selection and testing) algorithm

Feature	Class	Statistics	Time point	Number of features
Energy	GLCM	SD	T0	3
Inversion different moment	GLCM	SD	T0
Run length nonuniformity	GLRLM	SD	T0
Entropy	GLCM	SD	T1	5
Long run emphasis	GLRLM	Mean	T1
Inversion different moment	GLCM	SD	T1
Cluster shade	GLCM	Mean	T1
Long run high grey level emphasis	GLRLM	Mean	T1
Entropy	GLCM	Mean	T2	12
Cluster shade	GLCM	Mean	T2
Short run emphasis	GLRLM	SD	T2
Short run low grey level emphasis	GLRLM	SD	T2
Inertia	GLCM	SD	T2
Cluster shade	GLCM	SD	T2
Short run emphasis	GLRLM	SD	T2
Long run emphasis	GLRLM	SD	T2
Run length non-uniformity	GLRLM	Mean	T2
Run length non-uniformity	GLRLM	SD	T2
Short run low grey level emphasis	GLRLM	SD	T2
Long run low grey level emphasis	GLRLM	Mean	T2
Variance	Intensity		T3	8
Short run emphasis	GLRLM	SD	T3
Run length non-uniformity	GLRLM	Mean	T3
Low grey level run emphasis	GLRLM	Mean	T3
Short run high grey level emphasis	GLRLM	Mean	T3
Long run low grey level emphasis	GLRLM	SD	T3
Max	Intensity		T3
Inertia	GLCM	Mean	T3
Integrated intensity	Intensity		T4	7
Cluster prominence	GLCM	SD	T4
Grey level non-uniformity	GLRLM	Mean	T4
Short run high grey level emphasis	GLRLM	SD	T4
Long run low grey level emphasis	GLRLM	SD	T4
Mean	Intensity		T4
Long run emphasis	GLRLM	SD	T4

T0, T1, T2, T3, and T4 represent the time-points of the dynamic series when the features were selected; the number represents the quantity of features selected for each time-point
TWIST Training with input selection and testing, GLCM Grey level co-occurrence matrix, GLRLM Grey level run length matrix, SD Standard deviation