Christian Waldenberg - Development of new MRl-based analysis methods for improved diagnosis of low back pain patients

Link directly to the doctoral thesis in GUPEA

Abstract

Image

Background. Low back pain (LBP) is the leading cause of disability worldwide, where three of four individuals experience back pain at some point in their lives. The pathophysiological background of LBP is probably multifactorial, where bone marrow damage, tissue changes in the vertebral endplates, and intervertebral disc (IVD) degeneration have been recognized as tissue changes linked to pain. Annular fissures within the IVD are of particular interest, as they may be associated with vascular and nerve ingrowth. However, these tissue changes are also common in asymptomatic individuals, making it difficult to correctly identify the cause of pain in the individual patient

Aim. This thesis aims to develop data-driven MRI-based analysis methods to improve the understanding of spinal tissue changes and their association with LBP with the ultimate purpose of improving diagnostics.

Paper I. MR images of 49 IVDs in 10 LBP patients were analyzed with unsupervised clustering methods to objectively and continuously classify the IVD heterogeneity related to degenerative changes. IVD degeneration could successfully be quantified with the proposed method.

Paper II. The lumbar IVDs of 25 LBP patients and 12 matched controls were examined with T2 mapping to quantify possible differences in IVD signal behaviors between patients and controls. The cohorts differed significantly in nucleus pulposus signal. A sub-analysis revealed that this signal difference was related to IVD fissures, visible as high-intensity zones at the outer part of the annulus fibrosus.

Paper III. Radiomics features were extracted from 123 IVDs (n=43 LBP patients) and examined with conventional MRI followed by discography and computed tomography. The features were further analyzed using artificial neural networks and a radiomics-based attention mapping technique to identify the presence and position of possible annular fissures. The method showed great potential and was found to classify the presence of fissures with 100% sensitivity and 97% specificity. The method also identified the position of fissures in 87% of the analyzed IVDs.

Paper IV. Radiomic features from 61 LBP patients examined with conventional MR imaging were extracted and then analyzed using machine-learning techniques to explore possible associations between annular fissures and vertebral lesions. The findings suggest that radiomics can objectively detect vertebral tissue changes associated with adjacent annular fissures.

Conclusion. With data-driven methods, such as radiomics and attention mapping, tissue changes both within the IVD and the vertebra were well revealed in LBP patients. Further, the methods could be used to find associations between different types of tissue changes and were sensitive to subtle and imperceptible changes associated with disc degeneration and annular fissuring. These analysis methods could contribute to improved MRI diagnostics for LBP patients.