Shaofeng Jiang

Local Morphology Fitting Active Contour for Automatic Vascular Segmentation

In this paper, we propose an active contour model using local morphology fitting for automatic vascular segmentation on 2-D angiogram. The vessel and background are fitted to fuzzy morphology maximum and minimum opening, separately, using linear structuring element with adaptive scale and orientation. The minimization of the energy associated with the active contour model is implemented within a level set framework. As in the current local model, fitting the image to local region information makes the model robust against the inhomogeneous background. Moreover, selective local estimations for fitting that are precomputed instead of updated in each contour evolution makes the evolution of level set robust again initial location compared to the current local model. The results on synthetic image and real angiogram compared with other methods are presented. It is shown that the proposed method can achieve automatic and accurate segmentation of vascular angiogram.

Morphological Multiscale Enhancement, Fuzzy Filter and Watershed for Vascular Tree Extraction in Angiogram

This paper presented an automatic morphological method to extract a vascular tree using an angiogram. Under the assumption that vessels are connected in a local linear pattern in a noisy environment, the algorithm decomposes the vessel extraction problem into several consecutive morphological operators, aiming to characterize and distinguish different patterns on the angiogram: background, approximate vessel region and the boundary. It started with a contrast enhancement and background suppression process implemented by subtracting the background from the original angiogram. The background was estimated using multiscale morphology opening operators by varying the size of structuring element on each pixel. Subsequently, the algorithm simplified the enhanced angiogram with a combined fuzzy morphological opening operation, with linear rotating structuring element, in order to fit the vessel pattern. This filtering process was then followed by simply setting a threshold to produce approximate vessel region. Finally, the vessel boundaries were detected using watershed techniques with the obtained approximate vessel centerline, thinned result of the obtained vessel region, as prior marker for vessel structure. Experimental results using clinical digitized vascular angiogram and some comparative performance of the proposed algorithm were reported.

Automatic Segmentation of Cerebral Computerized Tomography Based on Parameter-Limited Gaussian Mixture Model

This paper realizes a new method to segment intracranial structure from series of cerebral computerized tomography (CT) automatically. Firstly, a region growing and morphology based approach is developed to extract intracranial structures from series cerebral computerized tomography with the knowledge of anatomy, and then focusing on the problems of parameter initialization of the expectation maximization (EM) algorithm, an improved EM algorithm based on Parameter- Limited Gaussian Mixture Model is presented to segment intracranial structures successfully. Experiment shows that this method is successful on all cerebral computerized tomography from bottom to top part of cerebra.

Segmentation of Coronary Artery on Angiogram by Combined Morphological Operations and Watershed

In this paper, a scheme using fuzzy mathematical morphology operations is proposed for extracting coronary ar- teries tree on angiogram. After the enhancement of image with the traditional morphological top-hat operator, the vessel image is filtered with the fuzzy morphological opening with a set of linear structuring elements at different orientation. At the same time, the enhanced image is also filtered with the dual of fuzzy opening with the same structuring elements set. The two filtered vessel image is then combined into a new image considering the local orientation consistency, detected by the morphological filter, within a linear neighborhood of each location. Threshold of the result image produces the binary vessel structure which is separate from background structure. The extracted vessel structure is lastly treated as prior of shape and location of vessel and used as marker in morphological watershed for detecting the accurate vessel boundaries. Experimentation shows the performance of proposed method on extraction vascular tree on clinical angiogram under x-ray. I. INTRODUCTION

Automatic Extraction of Brain from Cerebral MR Image Based on Improved BET Method

BET is a well known brain extraction method from cerebral MR, FMR and PET images, but there are cases that it can not get ideal result whatever you set the parameters in BET method when processing real cerebral MR images. To overcome this problem, this paper modifies the definitions of smoothing force and expansionary force used in the BET algorithm to evolve the contour of brain according to the intensity distribution of images and adds a new path to search the local maximum and minimum intensity of image. Experiments show that the improved method is more robust than BET method when processing real MR images.

Weighted Fuzzy Feature Matching for Region-Based Medical Image Retrieval: Application to Cerebral Hemorrhage Computerized Tomography

In this paper, we focus on retrieval for cerebral hemorrhage Computerized Tomography images based on Weighted Fuzzy Feature Matching (WFFM). We first apply an improved Expectation Maximization (EM) algorithm to segment the images into regions, and then extract the texture features of each region with Gabor filters. To improve the robustness of retrieval system against segmentation-related uncertainties, WFFM maps the intensity features of each region into fuzzy features with the exponential membership functions. Based on fuzzy features, regions between images are matched and the texture features serve as weighting factors when calculating the similarities between the images. Experiments show that the retrieval method performs better than some similar methods in the application to retrieve cerebral hemorrhage CT images.

Reweighting BiasMap Based Image Retrieval and Relevance Feedback for Medical Cerebral MRI Image

This paper proposed a region based image retrieval and relevant feedback (RF) system for Medical cerebral MRI images. In the system, firstly, the brains were extracted from cerebral images by a modified BET algorithm, and then were segmented into regions by EM algorithm based on Gauss Mixture Model. Each region was represented by fuzzy features. When performing retrieval, both regional and global features were used. To optimize the retrieval result, this paper used reweighting relevance feedback method (RW) to optimize regional features and proposed reweighting BiasMap based relevance feedback method (RW-BiasMap) to optimize global features. The computation of RW is very fast, but only uses the relevant images. RW-BiasMap is based on RW and BiasMap feedback method, it can use both the relevant images and the irrelevant images, but the computation of RW-BiasMap is slowly, so this paper only uses it to optimize the global features. Experiments show that this retrieval system is effective and RW-BiasMap performs better than BiasMap.

Content–Based Medical Image Retrieval Based on Fuzzy Image Structure and Content

This paper proposed a new method to retrieve cerebral hemorrhage CT images based on fuzzy binary tree structure and content (FBTS). Most of fuzzy-region-based CBIR systems only use fuzzy content information of images. FBTS uses both the fuzzy content features and the fuzzy structure features to retrieve images, and can merge the regions when some conditions are met. FBTS first segments the CT images into several regions with binary tree method, then gets the fuzzy binary tree structures by assigning each region a membership degree which the pixels belong to the region according to the intensity standard deviation of each region. The membership degree of each region servers as the weighting factor when calculating the similarity between images based on UFM method. Experiments show that FBTS is robust to the uncertainty of image segmentation.