Su Zhang

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.

SIFT and shape information incorporated into fluid model for non-rigid registration of ultrasound images

Non-rigid registration of ultrasound images takes an important role in image-guided radiotherapy and surgery. Intensity-based method is popular in non-rigid registration, but it is sensitive to intensity variations and has problems with matching small structure features for the existence of speckles in ultrasound images. In this paper, we develop a new algorithm integrating the intensity and feature of ultrasound images. Both global shape information and local keypoint information extracted by scale invariant feature transform (SIFT) are incorporated into intensity similarity measure as the body force of viscous fluid model in a Bayesian framework. Experiments were performed on synthetic and clinical ultrasound images of breast and kidney. It is shown that shape and keypoint information significantly improves fluid model for non-rigid registration, especially for alignment of small structure features in accuracy.

Multi-Modality Medical Image registration Using Support Vector Machines

The registration of multi-modality medical images is an important tool in surgical application. We presented a method of computing different modality medical images registration of the same patient. It incorporates prior joint intensity distribution between the two imaging modalities based on registered training images. The prior joint intensity distribution is modeled by support vector machine. Results aligning CT/MR and PET/MR scans demonstrate that it can attain sub-voxel registration accuracy. Furthermore, it is a fast registration method because support vector machine solution is sparse

Shape symmetry analysis of breast tumors on ultrasound images

Shape characteristics of malignant and benign breast tumors are significantly different. In this paper, the reflective symmetry of breast tumor shapes on ultrasound images was investigated. A new reflective symmetry measure (RSML) derived from multiscale local area integral invariant was proposed to quantify the shape symmetry of breast tumor, which could be computed directly from the binary mask image without the shape parameterization in terms of arc length. The performance of several symmetry measures for differentiating malignant and benign breast tumors at varying scales was evaluated and compared by receiver operating characteristic (ROC) analysis. RSML with Gaussian kernel at scale 0.04 (related to the maximal diameter) achieved the highest area under the ROC curve (0.85) on the image data of 168 tumors (104 benign and 64 malignant). The experimental results showed that the reflective symmetry of breast tumor shape was capable of providing potential diagnostic information, which could be characterized quantitatively by RSML with the appropriate scale parameter.

SVM for density estimation and application to medical image segmentation

A method of medical image segmentation based on support vector machine (SVM) for density estimation is presented. We used this estimator to construct a prior model of the image intensity and curvature profile of the structure from training images. When segmenting a novel image similar to the training images, the technique of narrow level set method is used. The higher dimensional surface evolution metric is defined by the prior model instead of by energy minimization function. This method offers several advantages. First, SVM for density estimation is consistent and its solution is sparse. Second, compared to the traditional level set methods, this method incorporates shape information on the object to be segmented into the segmentation process. Segmentation results are demonstrated on synthetic images, MR images and ultrasonic images.

Non-rigid Registration Algorithm Based on Modally Controlled Free Form Deformation

To register pre-operative MRI/CT images with intra-operative ultrasound images based on the vessels visible in both of the modalities, we presents a non-rigid registration method of multimodal medical images based on free form deformation was proposed. When the images are aligned, the centerline points of the vessels in one image will align with the intensity ridge points in the other image. Rigid transformation was adopted in global registration while local deformation was described by a free form deformation (FFD) based on a modally controlled B-spline. The method applies an optimization strategy combining the genetic algorithm with the conjugated gradients algorithm to minimize the objective function. Two experiments were designed on phantom and clinical data to evaluate the method. The results indicate that the registration method is consistent and suggest that it is accurate. The average standard deviation of the final transformation parameters is sub-voxel, sub-millimeter, and within 0.010 radians. The results show that the method has good registration accuracy and convergence rate. And it can be applied efficiently in the ultrasound-image-guided surgery system

Investigation of a Phenylalanine-Biosensor System for Phenylketonuria Detection

Detection and prevention of Phenylketonuria (PKU) is becoming more and more important. However, the current methods are either imprecise or time-consuming. We propose a biosensor system based on phenylalanine ammonia-lyase (PAL) immobilized on an ammonia electrode to measure blood phenylalanine for PKU prevention. The biosensor exhibits good linearity from 10-5000 muM and the response time is only about 2 minutes. It remains stable for at least 5 days and less than 20% drop of the original activity after ten day storage at 4 D, while the service life of the biosensor could be up to 30 days. We also develop an intelligent system to ensure optimal conditions for operation and preservation of the biosensor and to make detection more convenient and reliable. All of these advantages indicate that the newly developed method could be a better one for solving the problems of PKU detection

A Stereo Thermographic System for Intra-operative Surgery

A stereo far-infrared (FIR) system based on the trigonometric parallax is presented in this paper for locating the distal arterial tree from the coronary artery bypass graft. The system can obtain the wide angle-of-view range information in the near distance by changing the optical path of one fixed uncooled FIR camera. Meanwhile, the speed of pixel movement on the FIR imaging plane is discussed for avoiding the problem on the imaging blur because the images are taken in sequence from a scanning mirror for the real-time monitor of the operation. Besides, the view range is also considered under the different system parameter conditions. After the structure parameters are simulated for evaluating the performance, the optimum system can be designed. This thermal imaging technique is inexpensive, noninvasive and feasible for intra-operative surgery

Effective Shape Measures in Malignant Risk Assessment for Breast Tumor on Sonography

Malignant and benign breast tumors have different shape characteristics associated with their growth ways on sonography. Through analyzing the tumor shape pattern on the clinical images and the experimental results, we find that the tumor shape can be characterized on three aspects: convexity, ellipticity, and symmetry. In this paper, the shape measures are quantified using the polygonal approximation, the fitting ellipse, and local area integral invariant of contour respectively. The performance of these shape measures is evaluated on a breast ultrasound image data of 87 cases (49 benign and 38 malignant). Two combined convexity measures, an elliptic compactness, and a new reflection symmetry measure from local area integral invariant among the shape measures are appropriate and effective for distinguishing malignant and benign tumors, and all of their area under ROC curve can reach 0.9. They are significantly different between benign and malignant tumors on sonography, and show potential for the malignant risk assessment.

Partial dependence of breast tumor malignancy on ultrasound image features derived from boosted trees

Various computerized features extracted from breast ultrasound images are useful in assessing the malignancy of breast tumors. However, the underlying relationship between the computerized features and tumor malignancy may not be linear in nature. We use the decision tree ensemble trained by the cost-sensitive boosting algorithm to approximate the target function for malignancy assessment and to reflect this relationship qualitatively. Partial dependence plots are employed to explore and visualize the effect of features on the output of the decision tree ensemble. In the experiments, 31 image features are extracted to quantify the sonographic characteristics of breast tumors. Patient age is used as an external feature because of its high clinical importance. The area under the receiver-operating characteristic curve of the tree ensembles can reach 0.95 with sensitivity of 0.95 (61/64) at the associated specificity 0.74 (77/104). The partial dependence plots of the four most important features are demonstrated to show the influence of the features on malignancy, and they are in accord with the empirical observations. The results can provide visual and qualitative references on the computerized image features for physicians, and can be useful for enhancing the interpretability of computer-aided diagnosis systems for breast ultrasound.