In tensor MRI, a set of k-space MRI data points is acquired that includes one or more k-space subsets of MRI data points. An object orientation (or spatial transformation) corresponding to each of the k-space subsets is determined. Because the object orientation (or spatial transformation) can differ from subset to subset, the overall set of k-space data can be inconsistent with respect to object orientation (or spatial transformation). This possible inconsistency can be addressed by providing a k-space tensor model that includes object orientation and/or spatial transformation information corresponding to each of the subsets. A tensor MRI image can be reconstructed from the set of k-space MRI data points by using the k-space tensor model to account for object orientation and/or spatial transformation.