Analysis of sagittal lumbar spine MRI images remains an important step in automated detection and diagnosis of lumbar spinal stenosis. There are numerous algorithms proposed in the literature that can measure the condition of lumbar intervertebral discs through analysis of the lumbar spine in the sagittal view. However, these algorithms rely on using suitable sagittal images as their inputs. Since an MRI data repository contains more than just these specific images, it is, therefore, necessary to employ an algorithm that can automatically select such images from the entire repository. In this paper, we demonstrate the application of an image classification method using deep convolutional neural networks for this purpose. Specifically, we use a pre-trained Inception-ResNet-v2 model and retrain it using two sets of T1-weighted and T2-weighted images. Through our experiment, we can conclude that this method can reach a performance level of 0.91 and 0.93 on the T1 and T2 datasets, respectively when measured using the accuracy, precision, recall, and f1-score metrics. We also show that the difference in performance between using the two modalities is statistically significant and using T2-weighted images is preferred over using T1weighted images.