Lin Jiangli

Edge Enhancement and Filtering of Medical Ultrasonic Images Using a Hybrid Method

A hybrid method based on anisotropic diffusion (AD) is proposed for ultrasound speckle suppression and edge enhancement. This method is designed to utilize the different denoising properties of three techniques: median filtering, improved AD filtering and isotropic diffusion filtering. The gradient matrix is analyzed, and thresholds are chosen by experiments. The hybrid method is made by combining the three filtering methods for three different grayscale gradient ranges respectively. The filtering program is realized by iteration, with iteration time determined by iteration stopping criterion (ISC) accurately. In the experiment hundreds of images are processed by hybrid method, in contrast with median filtering and speckle reducing anisotropic diffusion (SRAD) filtering. The experimental results shows hybrid method can greatly improve processing speed, while successfully suppress speckle and enhance edge, therefore suitable for high-speed noise elimination of 3D images with huge data.

Segmentation of Medical Ultrasound Image Based on Markov Random Field

Segmentation is a most important but difficult step in ultrasound image analysis. For the speckle noise and the tissue intensity inhomogeneities in the medical ultrasound images, the conventional segmentation approaches based on intensity or intensity-statistics do not work well. Current studies to reduce the speckle noise are failed in boundary preserving. And the researches on intensity inhomogeneites can not obtain the complete structure. In this paper, a new segmental method combined Markov random field (MRF) model with morphological image processing is proposed to cover the shortages above. MRF step is used to estimate the label image and morphological image processing makes the region-of-interest (ROI) complete to get a complete tissue. This algorithm is insensitive to speckle noise. Experimental results on synthetic images and ultrasound images show that this algorithm works successfully in MRF model and can correctly identify the tissues in the medical ultrasound images.

Boundary-Based Feature Extraction and Recognition of Breast Tumors Using Support Vector Machine

Breast cancer is the most common cancer among women. To assist the ultrasound (US) diagnosis of solid breast tumors, the lobulated contour feature quantified by boundary-based corner counts is studied to classify breast tumors as malignant or benign. The corner points in this research was detected based on wavelet transform (WT), and the classification selected through comparison is Support Vector Machine (SVM), with radial based function (RBF) as the kernel function. Experiments were done on a total of 240 cases of breast lesions, including 104 cases of malignant tumors proved at histology and 136 cases of benign tumors. The accuracy of this system is 95.42%, specificity is 98.53% while sensitivity is 91.35%. Consequently, by SVM, the obtained results show that the pro-posed method can be a new intelligent assistance diagnosis.

Automatic 3D Segmentation of MRI Brain Images Based on Fuzzy Connectedness

An automatic 3D segmentation based on fuzzy connectedness (FQ) is proposed for MRI brain images. The main contribution of the present paper includes two parts: the accurate extraction of brain tissues and the automatic selection of the seed of FC. The brain tissues are extracted through obtaining approximate region of brain tissues via improved region growing, choosing the optimal threshold value to separate the background and brain tissues, eroding to disconnect duramater from the region of brain tissues, and blurring brain tissues with a smoothing linear filter to obtain the template of brain tissues. For automatic selection of the seed, according to the largest gray probability density of white matter and grey matter respectively corresponding to different gray-scale values, approximate regions of white matter and grey matter are estimated; in such approximate regions, the probability density of the volume data and the intensity uniformity are utilized to select the seed automatically. This method requires no user interaction, and is fully automatic and robust. The experimental results show that the proposed method can accurately select seeds and get accurate segmentation results.

Quantification of Eccentric Mitral Regurgitation by Regurgitant Volume

To investigate and verify that real-time 3-dimensional (RT-3D) color Doppler is capable of quantifying the eccentric miral regurgitation (MR) by computing regurgitant volume (RV). 34 patients (mean age 37.13 plusmn 18.28 years) with confirmed eccentric MR were undergone RT-3D echocardiography. By the spatial and temporal integration of the cross-sectional velocity distribution of regurgitant valve, the systolic RV can be evaluated. Then RV were applied to assessing the severity of eccentric MR and compared with the jet volume (JV) and the vena contracta width (VCW) immediately measured. In 34 patients, 2 patients were graded as mild, 16 patients as moderate, and 16 patients as severe. Our RV measurements have good correlation with JV(r=0.9453) and VCW(r=0.8829) respectively. RT-3D color Doppler echocardiography was capable of quantifying the RV of the eccentric MR.