An image analysis framework based on vertebrae segmentation in lumbar MRI is presented in this paper. As PCA lacks for describing data in detail

it is replaced by ICA in both shape and texture modeling with a further fast sorting of Independent Components. Improved ICA-AAMs is then built

with which vertebrae in MRI are segmented. Based on vertebra segmentation

the best fitting function of lumbar curve is obtained with the MDL principle; lumbar lordosis angles are then computed based on the curvature of this lumbar curve. At last

disc degeneration is evaluated from a declining intensity projecting profile between the endplates of vertebrae. Experiments have testified the accuracy and validity of the segmentation and quantitative analysis respectively. This framework reduces the clinical trifle in terms of providing impersonal quantitative information to doctors. It is also extensible to lumbar computer aided diagnosis (CAD).