Effect of slice thickness on brain magnetic resonance image texture analysis

Table 3 The accuracy of NDA method for T1 and T1C

ROIs	1 mm -> 1 mm		1 mm -> 3 mm		3 mm -> 1 mm		3 mm -> 3 mm
	T1	T1C	T1	T1C	T1	T1C	T1	T1C
WM vs. NAWM	59 [57, 60]	50 [48, 52]	54 [52, 56]	48 [46, 50]	54 [54, 55]	50 [50, 53]	57 [55, 58]	60 [59, 63]
WM vs. MSi	95 [94, 96]	96 [94, 97]	94 [93, 95]	92 [91, 93]	92 [91. 93]	88 [88, 89]	96 [96, 98]	93 [92, 94]
WM vs. MSr	92 [92, 93]	92 [91, 93]	85 [85, 86]	92 [91, 92]	90 [89, 90]	81 [80, 82]	88 [88, 89]	92 [91, 92]
NAWM vs. MSi	92 [91, 92]	98 [96, 98]	91 [90, 92]	97 [96, 98]	91 [90, 93]	98 [97, 98]	95 [93, 96]	98 [98, 100]
NAWM vs. MSr	90 [89, 90]	88 [88, 89]	86 [85, 88]	90 [89, 91]	89 [88, 90]	90 [89, 91]	86 [85, 86]	92 [91, 92]

Tissue classification with NDA. Results are given in percentages for T1 and T1C (with contrast agent) images, in comparisons of white matter (WM), normal appearing white matter (NAWM), MS plaques with irregular ROI (MSi) and MS plaques with regular ROI (MSr). Each comparison includes three texture parameters. Texture parameters have been chosen for NDA based on the Fisher coefficient. In the top row, "X -> Y" refers to the case in which the training set consists of slices with thickness X (1 mm/3 mm), whereas the test set comprises of slices with thickness Y (1 mm/3 mm). The 95% confidence interval for the median is shown in square brackets.

ISSN: 1475-925X