Table 2.

Performance on the testing set obtained by the single classifier

Feature		Formula
Morphological	Area, A k	n k dxdy where n k is number of voxels in the kth slice of the ROI, d x and d y represent size of voxels
	Perimeter length, P k	b k where b k is number of boundary voxels in the kth slice of the ROI
	Compactness in 3D, COMP	$\frac{S^2}{V_{3D}}$ where S is the surface and V is the volume, defined as follows: $s={\displaystyle \sum _x}{\displaystyle \sum _y}{\displaystyle \sum _z}{b}_{ROI}\left(x,y,z\right)v_{\mathit{size}}\mathit{slic}{e}_{th}$ where v size is the voxel size, slice th is the slice thickness; and V = n ROI dxdydz where n ROI is total number of voxels in the ROI
	Eccentricity, ECC	$\frac{\sqrt{a^2-b^2}}{a}$ where a is the major axle shaft and b is the lower one
Dynamic	Basal signal, BS	Signal intensity before contrast injection
	Relative enhancement, RE(t i)	$\frac{SI\left(t_i\right)-BS}{BS}$ where t i is the ith temporal instant
	Sum of local differences, SOD	$\mathit{SO}{D}_p=\mathit{Pr}{e}_p+{\displaystyle \sum _{i=1}^T}|Pos{t}_p\left(i\right)-Pos{t}_p\left(i-1\right)|$

ROI region of interest