The invention discloses a multidimensional vein extracting method based on a brain nuclear magnetic resonance image. An area-of-interests in the brain nuclear magnetic resonance image is segmented through a region growing method. Vein characteristic parameters of the area-of-interests are extracted through a Curvelet conversion and Contourlet conversion method. The people comprise an Alzheimer patient group, a mild cognitive impairment patient group and normal old people group, the vein characteristic parameters of the area-of-interests comprise entropy, gray average, correlation, energy, the homogeneity degree, variance, the maximum probability, deficit distance, an inverse gap, clustering tendency, contrast, a sum mean value, a difference mean value, a sum entropy and a difference entropy. The area-of-interests comprises a entorhinal cortex are and a sea horse area.