Predicting Mismatch-Repair Status in Rectal Cancer Using Multiparametric MRI-Based Radiomics Models: A Preliminary Study

Table 2

Comparisons of selected features between different MMR status.

No.	Radiomics feature	Sequence	dMMR median (interquartile range)	pMMR median (interquartile range)	value^†	value	Coefficients
1	Wavelet-LLH_ GLSZM_ zone variance	DWI	5.131 (4.378-5.300)	3.388 (2.950-3.968)	3.521	<0.001	-0.075
2	Wavelet-HLH_ GLSZM_ zone variance	DWI	6.244 (5.800-8.000)	5.000 (4.350-5.200)	2.312	0.021	-0.059
3	Wavelet-HLH_ GLSZM_ large area high gray level emphasis	CE-T1WI	4458.739 (3441.205-5718.022)	2524.119 (2098.914-3682.686)	2.505	0.012	-0.052
4	Gradient_ first order_ kurtosis	T2WI	175982.000 (92672.000-213782.500)	85442.833 (31509.800-107171.778)	2.955	0.003	-0.032

GLSZM: gray level size zone matrix. The wavelet transform decomposes the tumor area image into low-frequency components (L) or high-frequency components (H) in the , , and axes. ^†Mann–Whitney test. The coefficients in LASSO algorithm.