| Summary: | 碩士 === 中華大學 === 電機工程學系 === 106 === Cerebral microbleeds (CMBs) are products of local chronic blood on normal brain tissue, and the relation between CMBs and neurological diseases has been verified. The presence and quantity of CMBs in the brain are important for brain diagnosis. Physicians calculate the number and positions of micro bleeding points by means of diagnosing MRI images. The purpose of this research is to develop a computer aided detecting system which can precisely determine the CMB number and positions and reduce the time of diagnosis to assist physicians on patient observation and treatment.
In the thesis we propose a method using image processing and texture analysis techniques to construct a computer-aided diagnosis system for detecting CMBs. First, the fully convolutional network (FCN) is applied to filter out all the CMB candidates from the MRI data. Second, choose an appropriate size of block for each CMB candidate. Third, calculate the 3-D discrete wavelet transform (DWT) for each CMB candidate block. Fourth, extract features form the lowest frequency band of the 3D DWT coefficients of each CMB candidate. Finally, the support vector machine (SVM) is used for training and testing.
The MRI data of 20 patients were used in the research, and each subject has 150 512-by-512 MRI images. The total number of CMBs of the 20 patients is 71.
|
|---|
