| Summary: | 碩士 === 國立雲林科技大學 === 工業工程與管理系 === 106 === This study make an approach to spontaneous intracranial hypotension (SIH). In the past for the assessment of SIH, mostly qualitative assessment, but the qualitative assessment is too subjective. The advantage of direct quantitative assessment is that it can be knew the volume and leakage location.
The purpose of this study was to quantitatively assess cerebrospinal fluid by using image processing techniques. Since the magnetic resonance image of this study has unknown background characteristics and the intensity of cerebrospinal fluid is not uniform, it will affect the performance of the image segmentation algorithm. Therefore, image preprocessing must be used for correction. Image preprocessing uses the following five methods for comparison: (1) Non-Local mean (2) Adaptive Gamma Correction (3) Non-Local mean + Adaptive Gamma Correction (4) Adaptive Gamma Correction+ Non-Local mean (5) without pre-processing. The image segmentation process uses the Marker-Controlled Watershed algorithm in the regional model and the K-means, Otsu, and Global Entropy algorithms in the statistical model for comparison. Finally, Jaccard Similarity is used as a performance measure to compare the results of segment with the standard images.
This study was divided into magnetic resonance images of the spinal cord and magnetic resonance images of the brain. Among them, the magnetic resonance images of the spinal cord were taken from the magnetic resonance images provided by the Radiology Department of Veterans General Hospital of Taichung; since the standard images of brain's cerebrospinal fluid is difficulties in drawing.. This study will be simulated using images from the Brainweb’s database.The results of the image segmentation part are as follows: compared to Otsu、Marker-Controlled Watershed、Global Entropy, K-means algorithm can provide efficient and stable results in the study of spinal cord imaging, brain simulation images. Magnetic resonance imaging of the spinal cord showed a K-means algorithm with a mean performance of 0.9108 in a total of 8019 axial images, and a K-means algorithm with a mean performance of 0.9788 in 2610 axial images.
|
|---|
