Over the last century, Radiology departments have experienced rapid transformative evolution, driven by engineering innovations in medical imaging. The integration of Artificial Intelligence (AI)-based algorithms in medical imaging has started, with the aim of helping clinicians in addressing high workload challenges, mainly due to aging population. In this setting, the introduction of weight-bearing MRI give the opportunity to perform MRI scans of the lumbar spine in both supine and upright positions. This additional information has enhanced the detection of disk protrusion and translational intervertebral movement in degenerative spine diseases. This thesis focuses on developing and validating advanced medical imaging techniques for analyzing weight-bearing MRI images of the lumbar spine in normal subjects and patients with degenerative spine diseases. The proposed applications include the use of 3D image processing algorithms, 3D image reconstruction, and Deep Learning to provide clinically relevant information for the clinical assessment of patients. Furthermore, the algorithms are validated in two clinical studies. In Study 1, a 3D Convolutional Neural Network (CNN) segmentation network was specifically designed to identify and segment each vertebra and dural sac to obtain clinically relevant information related to spinal stenosis patients. Redundant nerve root is a well known sign associated with lumbar spinal canal stenosis (LSCS). LSCS subjects exhibited a significant decrease in antero-posterior cauda equina nerve roots dispersion cranially to the stenosis level and an increased antero-posterior dispersion caudally to the stenosis, as well as less homogeneity compared to normal subjects. In Study 2, the same segmentation network was applied to examine 3D vertebral micro-instability in spondylolisthesis patients in terms of vertebral center of mass, Meyerding grade, and roto-translation matrix between reference and slippery vertebrae. The analysis of the roto-translation matrix quantified 3D micromovements of vertebral volume. Statistical significance was evident in the antero-posterior direction at the L2-L3 level (p = 0.038) and cranio-caudal direction at the L3-L4 level (p = 0.006). These findings could serve as a foundation for addressing vertebral instability issues using 3D evaluation. In summary, this thesis advances the application of advanced medical imaging techniques for examining weight-bearing MRI images of the lumbar spine. The successful application of these techniques in degenerative spine diseases has revealed significant correlations between clinical information, uncovering promising new clinical patterns that have the potential to enhance diagnostic accuracy.
Development and Clinical Validation of Artificial Intelligence-based Software For Lumbar Weight-Bearing MRI / Levi, Riccardo. - (2024 Feb 28).
Development and Clinical Validation of Artificial Intelligence-based Software For Lumbar Weight-Bearing MRI
Levi, Riccardo
2024-02-28
Abstract
Over the last century, Radiology departments have experienced rapid transformative evolution, driven by engineering innovations in medical imaging. The integration of Artificial Intelligence (AI)-based algorithms in medical imaging has started, with the aim of helping clinicians in addressing high workload challenges, mainly due to aging population. In this setting, the introduction of weight-bearing MRI give the opportunity to perform MRI scans of the lumbar spine in both supine and upright positions. This additional information has enhanced the detection of disk protrusion and translational intervertebral movement in degenerative spine diseases. This thesis focuses on developing and validating advanced medical imaging techniques for analyzing weight-bearing MRI images of the lumbar spine in normal subjects and patients with degenerative spine diseases. The proposed applications include the use of 3D image processing algorithms, 3D image reconstruction, and Deep Learning to provide clinically relevant information for the clinical assessment of patients. Furthermore, the algorithms are validated in two clinical studies. In Study 1, a 3D Convolutional Neural Network (CNN) segmentation network was specifically designed to identify and segment each vertebra and dural sac to obtain clinically relevant information related to spinal stenosis patients. Redundant nerve root is a well known sign associated with lumbar spinal canal stenosis (LSCS). LSCS subjects exhibited a significant decrease in antero-posterior cauda equina nerve roots dispersion cranially to the stenosis level and an increased antero-posterior dispersion caudally to the stenosis, as well as less homogeneity compared to normal subjects. In Study 2, the same segmentation network was applied to examine 3D vertebral micro-instability in spondylolisthesis patients in terms of vertebral center of mass, Meyerding grade, and roto-translation matrix between reference and slippery vertebrae. The analysis of the roto-translation matrix quantified 3D micromovements of vertebral volume. Statistical significance was evident in the antero-posterior direction at the L2-L3 level (p = 0.038) and cranio-caudal direction at the L3-L4 level (p = 0.006). These findings could serve as a foundation for addressing vertebral instability issues using 3D evaluation. In summary, this thesis advances the application of advanced medical imaging techniques for examining weight-bearing MRI images of the lumbar spine. The successful application of these techniques in degenerative spine diseases has revealed significant correlations between clinical information, uncovering promising new clinical patterns that have the potential to enhance diagnostic accuracy.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.