Multiple linear stepwise regression of fiver lipid levels: proton MR spectroscopy study in vivo at 3.0 T

Abstract

Objective To analyze the correlations between liver lipid level determined by liver 3.0 T 1H-MRS in vivo and influencing factors using multiple linear stepwise regression. Methods The prospective study of liver 1H-MRS was performed with 3.0 T system and eight-channel torso phased-array coils using PRESS sequence. Forty-four volunteers were enrolled in this study. Liver spectra were collected with a TR of 1500 ms ,TE of 30 ms, volume of interest of 2 cm ×2 cm ×2 cm, NSA of 64 times. The acquired raw proton MRS data were processed by using a software program SAGE. For each MRS measurement, using water as the internal reference, the amplitude of the lipid signal was normalized to the sum of the signal from lipid and water to obtain percentage lipid within the liver. The statistical description of height, weight, age and BMI, Line width and water suppression were recorded, and Pearson analysis was applied to test their relationships. Multiple linear stepwise regression was used to set the statistical model for the prediction of Liver lipid content. Results Age (39.1 ± 12. 6) years, body weight (64.4 ± 10. 4) kg,BMI (23.3 ±3.1) kg/m2, linewidth (18.9 ±4.4) and the water suppression (90.7 ±6.5)% had significant correlation with liver lipid content (0.00 to 0.96%, median 0. 02% ), r were 0.11,0. 44,0. 40,0. 52, - 0. 73 respectively(P < 0. 05 ). But only age, BMI, line width, and the water suppression entered into the multiple linear regression equation. Liver lipid content prediction equation was as follows: Y =1.395-(0.021 × water suppression) + (0.022 × BMI) + (0.014 × line width) - ( 0. 064 × age),and the coefficient of determination was 0.613, corrected coefficient of determination was 0.59. Conclusion The regression model fitted well, since the variables of age, BMI, width, and water suppression can explain about 60% of liver lipid content changes. Key words: Liver; Magnetic resonance spectroscopy