| Summary: | 碩士 === 國立陽明大學 === 生物醫學影像暨放射科學系 === 102 === Purpose: Carotid stenosis may cause ischemic stroke of the brain, causing severe clinical symptoms and complication. To understand the severity of cerebral ischemia, patients with carotid stenosis and early ischemic stroke often needs cerebral perfusion study. Identifying normal and ischemic brain parenchyma in the cerebral perfusion images is clinically important for the treatment of ischemic stroke and arterial stenosis patients. In the current computed tomographic perfusion technique, experienced neuro-radiologist is needed for reading and interpreting the results for making the diagnosis. The aim of this research is to provide a fast and accurate tool for identifying normal and abnormal brain parenchyma, and segmentation of gray and white matter of normal brain parenchyma.
Materials and methods: The pixel of CT perfusion was segmented using three different techniques after removal of bone, signal outside of the bone, blood vessels and cerebrospinal fluid. In this first technique, various perfusion parametric images were calculated. Scatter plots of two different parameters were made. The distributions of normal and abnormal brain parenchyma were segmented by drawing a line on the scatter plot. In the second technique, the independent component analysis was applied for segmenting normal brain. In the third technique, the normal brain was segmented by applying Otsu's method to the time-to-peak of the brain voxel.
Results: Although the first technique can be used to distinguish normal and abnormal sides, it was subjective to find a suitable line. For the second technique, the normal white matter was erroneously classified as abnormal brain matter. We found the third technique using the time-to-peak images was effective for segmenting normal and abnormal brain parenchyma. Then using the third technique, we can segment gray and white matter by applying local Otsu's method to the normal brain parenchyma.
Conclusion: For patients with stroke or carotid stenosis, normal brain can be effectively segmented by applying a threshold to the TTP images.
|
|---|
