Desarrollo de técnicas computacionales para la detección de cambios morfológicos cerebrales
posted on July 13, 2022


Resumen

Single Subject Voxel Based Morphometry (SS-VBM) is a methodology for the analysis of brain morphological changes using magnetic resonance imaging. Unlike classic VBM, this technique allows single case studies. However, its applicability is limited by the excessively large number of false positives that produces. For that reason, its configuration aspects have not been studied as exhaustively as in VBM. To address this issue, first, a SS-VBM study was conducted on healthy subjects to analyze false positives and the influence of control group and smoothing kernel size. Second, the sensitivity of SS-VBM was evaluated on subjects with simulated atrophies in different regions and sizes, analyzing the influence of the smoothing kernel. Third, the sensitivity and specificity of SS-VBM was evaluated on patients with hippocampal sclerosis. Fourth, a methodology based on machine learning was developed for the reduction of false positives. The configuration aspects in SS-VBM were shown to be a crucial factors for the detection performance of morphological changes. In addition, the performance limits of the methodology could be determined. Finally, it was shown that it is possible to reduce false positives while preserving true positives using machine learning.