Quantitative Diffusion Tensor Image Analysis: A Clinical Approach to Central Nervous System Injuries
Research output: Book/Report › Doctoral thesis › Collection of Articles
|Number of pages||80|
|Publication status||Published - 29 Mar 2019|
|Publication type||G5 Doctoral dissertation (article)|
|Name||Tampere University Dissertations|
Mild traumatic brain injury (mTBI) can be defined as a traumatically induced brain function disruption which, in most cases, is not detectable by conventional medical imaging. mTBIs are grievous ailments due to high occurrence and a lack of distinct quantitative diagnostic tools and biomarkers. This signifies that the diagnosis of mTBI is based on subjective clinical measures. Extensive research has been carried out in order to find a clear correlation between DTI derived quantitative metrics and the post-mTBI brain WM. No uniform evidence of absolute conditions of pathology or association with post-injury prognosis has yet been found. However, many previous studies report different correlations between DTI metrics and postinjury brain WM. Unfortunately, the observed changes vary between studies, and final conclusions on the effects of mTBI on brain WM have yet to be made. One source of variation is the incoherency of the analysis methods used in the assessment of mTBI patients. Additionally, the heterogeneity of the studied patient cohorts hinders the chances of drawing a generalisable clinical conclusion.
Our work aims to overcome the issues in quantitative mTBI analysis methods by introducing a simple yet robust automated analysis method for human brain WM analysis. Our research began by applying a novel third-party group level analysis method, tract-based spatial statistics (TBSS), to an mTBI patient sample. We tested the whole sample and several subgroups for abnormal WM, but the results were negative. It was also noted during the study that TBSS would not be a suitable tool for clinical mTBI diagnostic purposes as the method is not fully modifiable for the assessment of individual patients and involves an excessive amount of complex image data manipulation. An additional study of traumatic spinal cord injury (SCI) patients was successfully performed applying TBSS. We found widespread neurodegenerative changes in the post-SCI cerebral WM, but also signs of possible neuroplasticity. The results further confirmed the method’s applicability to neuropathologic conditions with homogeneous effects on the brain WM microstructure.
Based on our findings, we began to create an automated analysis method using a region of interest (ROI) approach. We utilised human brain atlas-based ROIs in the analysis, which were automatically registered to the analysed subjects. The procedure ensures that the subjects’ images are not heavily processed. This in turn minimises the bias caused by image manipulation. The patients were compared against a normal population DTI reference value model created from our control subjects. The preliminary normal population DTI model created for the purpose was a successful quantitative model of DTI reference values. The normal population model could be used in a variety of clinical applications if a large enough number of control subjects were introduced to the model. The normal model would be especially useful in support of mTBI diagnosis methods.
In summary, this thesis has three conclusions. First, we found no DTI measurable associations between WM integrity and acute mTBI when applying TBSS. Second, we found extensive WM changes in the post-SCI brain, which imply an ongoing neuroplastic process in addition to the initial SCI-induced changes. These cerebral WM changes were far more extensive than previously reported. Third, an automated quantitative DTI brain analysis method with prospective clinical applications was introduced. The sensitivity and specificity of the automated method is at an acceptable level when used in conjunction with our preliminary control population set. For clinical applicability, the method requires minor refinements to its usability. More importantly, the normal population model needs to be updated to clinical standards by increasing its statistical power. A large enough normal population data pool could be achieved through an MRI data collection scheme resembling that of a biobank data collection method. In addition, machine learning could be applied in future to create better statistical models for the analysis with more accurate model predicted DTI scalar values.