Based On The Isodata Algorithm And The Dynamic Curves Of Dce-mri Assessment Of Breast Tumor Detection

Breast cancer is a chief malignant tumor endangers women's health along with an increasingly trend of incidence. It has become essential to take into extraordinary account of its prevention and remedy. Compare to other imaging techniques, Magnetic Resonance Imaging (MRI) for clinical diagnosis of breast cancer has some its own advantages. However, there are many problems when the case comes into clinical practice. To handle with the problems and challenges for the current multifarious manual interpretation situation and CAD technology, this thesis proposes an improved ISODATA algorithm and fuzzy ISODATA algorithm specifically applied to detecting lesions in the breast MR images. Results of experiments on both simulation and measured data verifies the validity and accuracy of the two kinds of clustering algorithm, while the performances of the two are compared at the same time. The main creative results are summarized as follows:(1) Describes the noise characteristics of the MR images and analyzes the characteristics o f mean filter. Analyzes images before and after denoise by the peak signal to noise ratio an d normalized variance. And verifies the feasibility of MR images denoising.(2) After introduction of the original ISODATA algorithm principles and specific steps, proposes modified ISODATA algorithm improving its shortcomings specific in breast MR image clustering, and then employs clinical data validate the performance of the algorithm. And finally utilizes DCE-MRI kinetic curve to identify benign and malignant tumors.(3) Describes the theoretical basis of fuzzy ISODATA clustering analysis based on the obje ctive function, including data normalization methods, construct fuzzy similarity matrix and fuzz y classification methods. After that, exposition on the basis of the fuzzy ISODATA clustering a nalysis methods, and provides inspection standards of clustering. In addition, employ the mea sured data which is mentioned in the previous Chapter clustering, and utilizes kinetic curve t o identify benign and malignant tumors.