Ramlackhansingh, A; (2007) Comparison of methods for automated lesion identification in stroke patients. Doctoral thesis , UCL (University College London).

Preview

Text Ramlackhansingh_thesis.pdf Download (9MB) | Preview

Abstract

Introduction Lesion - Symptom mapping forms the foundation to our understanding of the function of different parts of the brain. As the science of neuro-imaging has developed, the detail and quality of images produced on scanning patients have improved. The result is that a single scan can produce a wealth of data. The manual analysis and interpretation of this data have thus become a time consuming and laborious affair. Manual analysis also has the drawback of being highly subjective and user dependent. This has resulted in the development of automated methods for the interpretation of such data. Before data could be analysed, it must be pre-processed into a format that can be optimally used by these methods. This step involves the normalisation, segmentation and spatial smoothing of patient scans. Scans must be pre- processed before automated analysis can be done. The use of automated methods is supposed to make data interpretation quick, reliable and reproducible. 1.2 Aim In this study the smoothing aspect of pre-processing will first be looked at. Here different smoothing levels will be used to try and decide on a value for optimal smoothing. Once this is decided analyses using three automated lesion identification methods will be done. The methods include Voxel Based Morphometry (VBM), Posterior Probability Mapping (PPM) and Fuzzy Clustering with fixed Prototypes (FCP). Parameters for the optimal functioning of these methods will be determined and suggestions will also be made as to which method may be the best and under what circumstances. 1.3 Results The optimal Full Width at Half Maximum (FWHM) value for smoothing was found to be 8mm. When the methods were looked at the following could be said: VBM: Analyses produced results in all 5 patients and Results only produced one level of information. PPM: The optimal probability threshold was determined to be 0.5 Analysis provides two levels of information. Lesions were identified with sharper borders and Analyses required the longest computer processing time. FCP: An a value of 0.5 provides the best results. Lesions identified with less sharp (i.e. fuzzy) borders - method possibly more sensitive and Analyses required the least computer processing time. 1.4 Conclusion This study provides further evidence that smoothing must be carried out on all scans to enable accurate and reliable lesion identification. An optimal FWHM of 8mm was determined. The study also determines an optimal value of 0.5 for the probability threshold in PPM analysis. An optimal a value of 0.5 was also found for FCP analysis. VBM proved to be the easiest method to use while PPM estimated the confidence to declare a tissue as abnormal and FCP was the most sensitive to lesion presence. Further work, however, should be done to further investigate the probability threshold for PPM and a value for FCP.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item