Xuesong Lu

Mutual information-based multimodal image registration using a novel joint histogram estimation

Mutual information (MI)-based image registration has been proved to be very effective in multimodal medical image applications. For computing the mutual information between two images, the joint histogram needs to be estimated. As we know, the joint histogram estimation through linear interpolation and partial volume (PV) interpolation methods may result in the emergency of the local extreme in mutual information registration function. The local extreme is likely to hamper the optimization process and influence the registration accuracy. In this paper, we present a novel joint histogram estimation method (HPV) by using an approximate function of Hanning windowed sinc as kernel function of partial volume interpolation. We apply it to both rigid registration and non-rigid registration. In addition, we give a new method estimating the gradient of mutual information with respect to the model parameters during non-rigid registration. By the experiments on both synthetic and real images, it is clearly shown that the new algorithm has the ability to reduce the local extreme, and the registration accuracy is improved.

Stick-guided lateral inhibition for enhancement of low-contrast image

The inhibitory interaction has long been observed in the lateral eye of the Limulus and been integrated into mechanism of enhancing contrast. When applying to the enhancement of low-contrast image for segmenting interested objects, the original lateral inhibition model will simultaneously amplify noises while enhancing edges contrast. This paper presents a new lateral inhibition model, which is called Stick-Guided Lateral Inhibition, for enhancement of low-contrast image so that week edges may exert a stronger force to catch the boundary of targets in the latter segmentation. First, the guided inhibition term is introduced as a general framework for improving the performance of lateral inhibition models in the presence of noises. Then, by using asymmetric sticks to guide the inhibiting process, we are able to accentuate the intensity gradients of image-edges and contours while suppressing the amplification of noises. Experiments on synthetic images and remote sensor images show that our model significantly enhances low-contrast images and improves the performance of latter segmentation.

Viscous fluid model-based non-rigid registration incorporating scale space keypoints

Non-rigid registration of monomodal image often takes an important role in image-guided radiotherapy and surgery. Viscous fluid model is widely used to enforce the topological properties on the deformation, and thus constrain the enormous solution space. Intensity-based method is popular in non-rigid registration, but it is sensitive to intensity variations. In this paper, we develop a new algorithm integrating the intensity and feature of images. The local keypoint information extracted by scale invariant feature transform (SIFT) in scale space is incorporated into intensity similarity measures as the body force of viscous fluid model in a Bayesian framework. Experiments were performed on synthetic and clinical images of CT and MR. It is shown that the keypoint information significantly improves fluid model for non-rigid registration in accuracy.

Integrated shape corresponding information by landmark sliding for non-rigid registration between CT and ultrasound images

Non-rigid registration between CT and ultrasound images is a difficult task due to the low resolution and contrast of ultrasound images. We present a method incorporating the shape information of contours extracted from the image pairs for the registration. Firstly, the shapes are represented by the automatically detected landmarks along the contour of segmented object on CT and ultrasound images. Secondly, the landmarks are corresponded by landmark sliding and the negative mean squared distance of the corresponding landmarks is used as the shape similarity measure. Finally, the integrated similarity metric for registration combines mutual information, shape similarity and smoothness constraint. The experiment results show that registration accuracy using the integrated similarity is higher than using only mutual information with nearly same execution time. It is verified that the shape information could improve the registration results between CT and ultrasound images.

Mutual Information-Based Multimodal Non-Rigid Image Registration Using Free-Form Deformation with A New Joint Histogram Estimation

In this paper, we present a voxel-based non-rigid registration algorithm using an improved interpolation method based on PV (partial volume) interpolation to estimate the joint histogram. Traditionally, the local deformation is modeled by a free-form deformation based on B-splines, and the cost function is the weighted combination of mutual information and a smooth constraint penalty term. In order to reduce the local extreme, we employ an approximate function of Hanning windowed sine as kernel function of PV interpolation, and use a new method computing the gradient of mutual information with respect to the model parameters efficiently. The experiment on CT abdomen images shows that the algorithm is easier to converge at the global optimum and restrain the local extreme.

A New Approach for Automatic Segmentation of LSCM Blood Vessel Images of Time Sequence Based on Region Growing

In the research of drug delivery, the ratio of the extravascular fluorescence intensity to the intravascular fluorescence intensity with respect to a transformed time axis plays an important role. Therefore, it is critical to accurately segment the blood vessel structures from time sequence images of LSCM. Conventionally, the users have to place some seed points into the images based on the region growing method. In this paper, the seed points are automatically taken from itself. The new approach also reduces the range of region growing in the images based on morphology and the computation speed becomes faster. In fact, it utilizes the correlative information between the images which are imaged at adjacent selected time-points during the time-course of experiment. Moreover, edge constraint is combined with region growing to reduce the influence of noisy boundaries and holes within the object. By conducting some experiments of live mice, it is demonstrated that the new approach is able to distinguish correctly all fluorescent pixels inside and outside the vessels, and the segmentation result is used to evaluate the permeability of drug.

Automatic Segmentation of Rat Mammary Glands from Serial MRI Images

A novel framework for automatic segmentation of rat mammary glands in MRI image sequences is presented in this paper. The Cartoon-Texture model is utilized in serial image segmentation to decompose the image into cartoon image and texture image. Then two-phase direct energy segmentation based on Chan-Vese active contour model is implemented on the cartoon image to partition the image into a set of regions. Seeds searching technology is applied iteratively on the texture image to find valid seeds for extracting the whole gland boundary points from the generated regions by a tracing algorithm we proposed. In iteration every time, texture images and features of the image patch around the seed are updated for new seeds searching and segmentation. Our segmentation approach does not require that the number of glands be identical or the location of the glands be close among consecutive images. Experiments show that our method is effective and efficient.

Automatic image registration by extraction of interested sub-region

With the recent increase in the amount of remote sensing images and the corresponding interest in temporal change detection, image registration has become increasingly important as a necessary first step in the integration of multi-temporal or multi-source data, for applications such as the analysis of seasonal and annual global climate changes, as well as land cover changes. Ground Control Point method can be subjective and extremely time-consuming as a result of manual selection of the control point pairs. The intensity-based registration is a solution to this problem. To the larger size images, however, the computational load can be high. In this paper, we develop a new automated registration approach to combine the advantage of two methods. First, a mathematical morphology-based method is used to extract interested sub-region pairs from the reference and floating image. Next, the registration of sub-region pairs is implemented by mutual information. At last the global transformation result will originate from the registration results of these sub-region pairs. By experiment on the larger size remote sensing images, less registration time and subpixel accuracy of registration results can be obtained, which reveal that the proposed registration algorithm is a robust solution.

A Wide Angle-of-View Thermal Imaging System Utilizing the Rotation of Optical Path

A wide angle-of-view far infrared (FIR) system based on the method of optical path rotation is presented. The system integrated one FIR camera and one rotating Gold-sprayed mirror controlled by one stepping motor. The FIR camera takes image sequence from the rotating mirror in the real-time monitor. Meanwhile, the speed of pixel movement on the FIR imaging plane is analyzed to obtain the rapidest data acquisition without motion blur. Besides, the view range of the proposed system is considered. Then, the prototype has been designed according to the results of simulation. The experiment data has verified the theory of motion blur. This thermal imaging technique is complexless, noninvasive and feasible for many fields involving the automatic temperature acquisition module.