| Title |
Selective Use of PoseNet Models to Improve Lumbar Vertebrae Detection Performance in X-ray Images |
| DOI |
https://doi.org/10.5573/ieie.2024.61.1.57 |
| Keywords |
X-lay image; Lumbar vertebrae; Artifical intelligence; Convolutional neural network; Localization |
| Abstract |
Recent advancements in AI technology have improved the performance of algorithms for automating medical image interpretation, including the analysis of X-ray images of the lumbar spine. The primary aim of these studies is to detect common lumbar diseases, such as spondylolisthesis and anterolisthesis. To achieve this goal, it is essential to detect firstly the positions of individual lumbar vertebra within the X-ray images. Various algorithms based on PoseNet have been proposed and utilized in different research projects for this purpose. However, challenges remain due to image ambiguity and variations in lumbar structures. In this study, we introduce both the conventional PoseNet and an additional modified model and utilize them selectively to propose a method for more accurately detecting the positions of the lumbar vertebrae in X-ray images of lateral view. To validate our proposed method, we employed 3,600 X-ray images and conducted ten simulations with different assignments of training, validation, and test data. The results consistently showed the improved performance of lumbar vertebrae detection compared to the conventional PoseNet in all simulations, reducing the average error rate to 0.28%. The proposed method can enhance the analysis performance of X-ray images for lateral lumbar spine and has potential applications in medical diagnostic support and automation systems. |