Deep learning system for Meyerding classification and segmental motion measurement in diagnosis of lumbar spondylolisthesis

Abstract

Lumbar spondylolisthesis, which is a common disorders caused by the extent of slipping of the spondylolysis vertebra and making the spinal posture unstable, mostly provides no painful symptoms and, consequently, can only be diagnosed by using X-rays. Manual methods currently used for measuring the slipping of the vertebra and segmental motions in X-rays, is not only considered as a high-levelled pre-anatomy process requiring high experienced surgeons to ensure the measurement accuracy but also practically ineffective for in handling a large number of X-ray images. Therefore, this paper mainly concerns the development of a deep learning system with supported by the convolutional neural network (CNN), in which the supplementary CNN model is trained to re-correct the keypoints located on corners of vertebra based on the first CNN regression model, to precisely measure required characteristics. In order to determine the instability of the lumbar spondylolisthesis, the measurement ability of the proposed method is also required to adapt to multiple lateral bending views including flexion and extension postures. Finally, the performance obtained has been validated with comparison between standard references measured from an experienced surgeon and automatic measured values which appreciably performed precise results with the mean deviation of 1.76° within 0.12 s for treating a single X-ray image on the computing configuration utilized.