Lumbar diseases are quite common and inevitable these days, due to the modern lifestyle in humans of almost all age groups. The main helpline is to get the disease diagnosed at the earliest possible with the advent of technology. Computer Assisted Diagnostic (CAD) system for disease detection and classification are quite rare for spine diseases in the existing literature, as it is difficult to process. This work proposes computer assisted diagnosis of lumbar spondylolisthesis that can differentiate among three classes namely no spondylolisthesis, mild and severe. The foreground detection is done using the grabcut algorithm to extract the spinal cord. The lumbar vertebrae L1 through L5 are extracted by marker-controlled watershed in association with the convex hull algorithm. The segmented images are described using both implicit and explicit feature extraction. In explicit feature extraction, various geometrical and textural features are taken using Grey Level Co-occurrence Matrix (GLCM). In implicit feature extraction, the transfer learning approach is used by utilizing the pre-trained model namely Visual Geometry Group network (VGG16) with fine tuning the customised fully connected layers. Support Vector Machine (SVM) multi class classifier is adapted to classify the images based on the explicitly extracted features and the softmax layer of VGG16 is used to classify the images based on the implicitly extracted features. The T1-weighted Magnetic Resonance Images (MRI) of 515 patients from the Lumbar Spine MRI Dataset in Mendeley Data website have been used for experimentation. The proposed model results in better performance with 94.10% of accuracy for implicit feature extraction and 92.44% of accuracy for explicit feature extraction as compared with existing active contour algorithm with 86.54% of accuracy for implicit feature extraction and 81.87% of accuracy for explicit feature extraction for the mentioned dataset.