Balerdi, G;
Henckel, J;
Di Laura, A;
Hart, A;
Belzunce, M;
(2024)
Automatic Segmentation Technique for Lumbar Spine Muscle Evaluation from MRI Images.
In: Ballina, FE and Armentano, R and Acevedo, RC and Meschino, GJ, (eds.)
Advances in Bioengineering and Clinical Engineering.
(pp. pp. 80-87).
Springer Nature
Preview |
Text
Baleardi.pdf - Accepted Version Download (574kB) | Preview |
Abstract
New quantitative muscle quality assessment tools are needed to improve the diagnosis, treatment, and study of the lumbar spine muscles. Quantitative magnetic resonance imaging (MRI) has become an important and increasingly relevant technique for diagnosing muscular diseases, tracking their progression, and measuring muscle composition. The Dixon sequence provides fat-only and water-only images, which allows the evaluation of muscle composition and size. Nevertheless, to discriminate a single muscle, a health professional has to manually segment the muscle from the MRI image, which is a slow and impractical process. In this study, we introduce a deep learning-based solution to automatically segment the lower spine muscles from Dixon MRI scans. To achieve that, we trained and validated a U-Net model using 26 manually segmented MRI images of the lower back muscles that was capable of automatically segmenting this muscle group, achieving a mean Dice score of 0.88 in the validation set. This high level of accuracy could allow the execution of new research looking at the size and composition of this muscle group and may also serve as a valuable tool for enhancing the diagnosis and treatment of lower back issues.
Archive Staff Only
 |
View Item |
