Renchao Jin

Segmentation of Lung Nodules in Computed Tomography Images Using Dynamic Programming and Multidirection Fusion Techniques1

RATIONALE AND OBJECTIVES The aim of this study was to develop a novel algorithm for segmenting lung nodules on three-dimensional (3D) computed tomographic images to improve the performance of computer-aided diagnosis (CAD) systems. MATERIALS AND METHODS The database used in this study consists of two data sets obtained from the Lung Imaging Database Consortium. The first data set, containing 23 nodules (22% irregular nodules, 13% nonsolid nodules, 17% nodules attached to other structures), was used for training. The second data set, containing 64 nodules (37% irregular nodules, 40% nonsolid nodules, 62% nodules attached to other structures), was used for testing. Two key techniques were developed in the segmentation algorithm: (1) a 3D extended dynamic programming model, with a newly defined internal cost function based on the information between adjacent slices, allowing parameters to be adapted to each slice, and (2) a multidirection fusion technique, which makes use of the complementary relationships among different directions to improve the final segmentation accuracy. The performance of this approach was evaluated by the overlap criterion, complemented by the true-positive fraction and the false-positive fraction criteria. RESULTS The mean values of the overlap, true-positive fraction, and false-positive fraction for the first data set achieved using the segmentation scheme were 66%, 75%, and 15%, respectively, and the corresponding values for the second data set were 58%, 71%, and 22%, respectively. CONCLUSION The experimental results indicate that this segmentation scheme can achieve better performance for nodule segmentation than two existing algorithms reported in the literature. The proposed 3D extended dynamic programming model is an effective way to segment sequential images of lung nodules. The proposed multidirection fusion technique is capable of reducing segmentation errors especially for no-nodule and near-end slices, thus resulting in better overall performance.

A novel fluence map optimization model incorporating leaf sequencing constraints

A novel fluence map optimization model incorporating leaf sequencing constraints is proposed to overcome the drawbacks of the current objective inside smoothing models. Instead of adding a smoothing item to the objective function, we add the total number of monitor unit (TNMU) requirement directly to the constraints which serves as an important factor to balance the fluence map optimization and leaf sequencing optimization process at the same time. Consequently, we formulate the fluence map optimization models for the trailing (left) leaf synchronized, leading (right) leaf synchronized and the interleaf motion constrained non-synchronized leaf sweeping schemes, respectively. In those schemes, the leaves are all swept unidirectionally from left to right. Each of those models is turned into a linear constrained quadratic programming model which can be solved effectively by the interior point method. Those new models are evaluated with two publicly available clinical treatment datasets including a head-neck case and a prostate case. As shown by the empirical results, our models perform much better in comparison with two recently emerged smoothing models (the total variance smoothing model and the quadratic smoothing model). For all three leaf sweeping schemes, our objective dose deviation functions increase much slower than those in the above two smoothing models with respect to the decreasing of the TNMU. While keeping plans in the similar conformity level, our new models gain much better performance on reducing TNMU.

A novel segmentation method for convex lesions based on dynamic programming with local intra-class variance

Lesion segmentation plays an important role in medical image processing and analysis. There exist several successful dynamic programming (DP) based segmentation methods for general images. In those methods, the gradient is used as an important factor in the cost function to attract the contours to the boundaries. Since medical images have their characteristics such as low contrast, blurred edges and high noises, the gradient operator cannot work well enough to achieve a satisfactory performance for boundary detection. We define the local intra-class variance and combine it with the dynamic programming method to replace the traditional gradient operation. Experiments on synthetic and X-ray images are carried out and the results are compared with Canny and fast multilevel fuzzy edge detection (FMFED) algorithms. It is demonstrated that the proposed method performs better on medical images with low contrast, blurred edges and high noises. In addition, 483 regions of interest of mammograms randomly extracted from DDSM of the University of South Florida are used to compare our proposed method with the plane-fitting and dynamic programming method (PFDP), and the normalized cut segmentation method (Ncut). The results demonstrate that our method is more accurate and robust than PFDP and Ncut.

Computer-aided detection of mammographic masses based on content-based image retrieval

A method for computer-aided detection (CAD) of mammographic masses is proposed and a prototype CAD system is presented. The method is based on content-based image retrieval (CBIR). A mammogram database containing 2000 mammographic regions is built in our prototype CBIR-CAD system. Every region of interested (ROI) in the database has known pathology. Specifically, there are 583 ROIs depicting biopsy-proven masses, and the rest 1417 ROIs are normal. Whenever a suspicious ROI is detected in a mammogram by a radiologist, it can be submitted as a query to this CBIRCAD system. As the query results, a series of similar ROI images together with their known pathology knowledge will be retrieved from the database and displayed in the screen in descending order of their similarities to the query ROI to help the radiologist to make the diagnosis decision. Furthermore, our CBIR-CAD system will output a decision index (DI) to quantitatively indicate the probability that the query ROI contains a mass. The DI is calculated by the query matches. In the querying process, 24 features are extracted from each ROI to form a 24-dimensional vector. Euclidean distance in the 24-dimensional feature vector space is applied to measure the similarities between ROIs. The prototype CBIR-CAD system is evaluated based on the leave-one-out sampling scheme. The experiment results showed that the system can achieve a receiver operating characteristic (ROC) area index AZ =0.84 for detection of mammographic masses, which is better than the best results achieved by the other known mass CAD systems.

Boundary Refined Texture Segmentation Based on K-Views and Datagram Methods

We propose a new texture segmentation algorithm to improve the segmentation of boundary areas in the image. In some applications such as medical image segmentation, an exact segmentation on the boundary areas is needed. But satisfactory segmentation results cannot be obtained on the boundary areas among different texture classes with some existing texture segmentation algorithms in our preliminary experiments. The proposed algorithm consists of three steps. The first step is to apply the K-view-datagram segmentation method to the image to obtain an initial segmentation; the second step is to find a boundary set which includes the pixels with high probabilities to be misclassified by the initial K-view-datagram segmentation; the third step is to apply a modified K-views template method with a small scanning window to the boundary set to refine the segmentation. The evaluation of the proposed algorithm was carried out with the benchmark images randomly taken from Brodatz Gallery and the ultrasonic prostate images provided by the hospitals. Initial experimental results show that the concept of boundary set defined in this paper can catch most of misclassified pixels of the output of the initial K-View-datagram segmentation. The new segmentation algorithm gives high segmentation accuracy and classifies the boundary areas better than the existing algorithms

Label fusion method based on sparse patch representation for the brain MRI image segmentation

The multi-Atlas patch-based label fusion method (MAS-PBM) has emerged as a promising technique for the magnetic resonance imaging (MRI) image segmentation. The state-of-the-art MAS-PBM approach measures the patch similarity between the target image and each atlas image using the features extracted from images intensity only. It is well known that each atlas consists of both MRI image and labelled image (which is also called the map). In other words, the map information is not used in calculating the similarity in the existing MAS-PBM. To improve the segmentation result, the authors propose an enhanced MAS-PBM in which the maps will be used for similarity measure. The first component of the proposed method is that an initial segmentation result (i.e. an appropriate map for the target) is obtained by using either the non-local-patch-based label fusion method (NPBM) or the sparse patch-based label fusion method (SPBM) based on the grey scales of patches. Then, the SPBM is applied again to obtain the finer segmentation based on the labels of patches. The authors called these two versions of the proposed fusion method as MAS-PBM-NPBM and MAS-PBM-SPBM. Experimental results show that more accurate segmentation results are achieved compared with those of the majority voting, NPBM, SPBM, STEPS and the hierarchical multi-atlas label fusion with multi-scale feature representation and label-specific patch partition.

Myocardial motion estimation of tagged cardiac magnetic resonance images using tag motion constraints and multi-level b-splines interpolation.

Myocardial motion estimation of tagged cardiac magnetic resonance (TCMR) images is of great significance in clinical diagnosis and the treatment of heart disease. Currently, the harmonic phase analysis method (HARP) and the local sine-wave modeling method (SinMod) have been proven as two state-of-the-art motion estimation methods for TCMR images, since they can directly obtain the inter-frame motion displacement vector field (MDVF) with high accuracy and fast speed. By comparison, SinMod has better performance over HARP in terms of displacement detection, noise and artifacts reduction. However, the SinMod method has some drawbacks: 1) it is unable to estimate local displacements larger than half of the tag spacing; 2) it has observable errors in tracking of tag motion; and 3) the estimated MDVF usually has large local errors. To overcome these problems, we present a novel motion estimation method in this study. The proposed method tracks the motion of tags and then estimates the dense MDVF by using the interpolation. In this new method, a parameter estimation procedure for global motion is applied to match tag intersections between different frames, ensuring specific kinds of large displacements being correctly estimated. In addition, a strategy of tag motion constraints is applied to eliminate most of errors produced by inter-frame tracking of tags and the multi-level b-splines approximation algorithm is utilized, so as to enhance the local continuity and accuracy of the final MDVF. In the estimation of the motion displacement, our proposed method can obtain a more accurate MDVF compared with the SinMod method and our method can overcome the drawbacks of the SinMod method. However, the motion estimation accuracy of our method depends on the accuracy of tag lines detection and our method has a higher time complexity.

Multiframe super-resolution based on half-quadratic prior with artifacts suppress

Involve the reconstruction properties of the AA prior model together with the Tikhonov prior model using a convex combination.By analyzing local features of the interesting image, the two prior models are combined by using an automatically calculated weighting function, making both smooth and discontinuous pixels handled properly.To demonstrate the efficiency of the combination strategy, the theoretical and practical analysis is presented in this paper.Our approach is formulated in framework of Bayesian statistics by the utilization of Kullback-Leibler divergence. Thus the motion parameters and hyper-parameters related to the composite prior and noise statistics are all determined automatically, leading to an unsupervised SR algorithm. The multiframe super-resolution (SR) technique aims to obtain a high-resolution (HR) image by using a set of observed low-resolution (LR) images. In the reconstruction process, artifacts may be possibly produced due to the noise, especially in presence of stronger noise. In order to suppress artifacts while preserving discontinuities of images, in this paper a multiframe SR method is proposed by involving the reconstruction properties of the half-quadratic prior model together with the quadratic prior model using a convex combination. Moreover, by analyzing local features of the underlined HR image, these two prior models are combined by using an automatically calculated weight function, making both smooth and discontinuous pixels handled properly. A variational Bayesian inference (VBF) based algorithm is designed to efficiently and effectively seek the solution of the proposed method. With the VBF framework, motion parameters and hyper-parameters are all determined automatically, leading to an unsupervised SR method. The efficiency of the hybrid prior model is demonstrated theoretically and practically, which shows that our SR method can obtain better results from LR images even with stronger noise. Extensive experiments on several visual data have demonstrated the efficacy and superior performance of the proposed algorithm, which can not only preserve image details but also suppress artifacts.

A Variational Bayesian Superresolution Approach Using Adaptive Image Prior Model

The objective of superresolution is to reconstruct a high-resolution image by using the information of a set of low-resolution images. Recently, the variational Bayesian superresolution approach has been widely used. However, these methods cannot preserve edges well while removing noises. For this reason, we propose a new image prior model and establish a Bayesian superresolution reconstruction algorithm. In the proposed prior model, the degree of interaction between pixels is adjusted adaptively by an adaptive norm, which is derived based on the local image features. Moreover, in this paper, a monotonically decreasing function is used to calculate and update the single parameter, which is used to control the severity of penalizing image gradients in the proposed prior model. Thus, the proposed prior model is adaptive to the local image features thoroughly. With the proposed prior model, the edge details are preserved and noises are reduced simultaneously. A variational Bayesian inference is employed in this paper, and the formulas for calculating all the variables including the HR image, motion parameters, and hyperparameters are derived. These variables are refined progressively in an iterative manner. Experimental results show that the proposed SR approach is very efficient when compared to existing approaches.

An improved label fusion approach with sparse patch-based representation for MRI brain image segmentation

The multi‐atlas patch‐based label fusion (LF) method mainly focuses on the measurement of the patch similarity which is the comparison between the atlas patch and the target patch. To enhance the LF performance, the distribution probability about the target can be used during the LF process. Hence, we consider two LF schemes: in the first scheme, we keep the results of the interpolation so that we can obtain the labels of the atlas with discrete values (between 0 and 1) instead of binary values in the label propagation. In doing so, each atlas can be treated as a probability atlas. Second, we introduce the distribution probability of the tissue (to be segmented) in the sparse patch‐based LF process. Based on the probability of the tissue and sparse patch‐based representation, we propose three different LF methods which are called LF‐Method‐1, LF‐Method‐2, and LF‐Method‐3. In addition, an automated estimation method about the distribution probability of the tissue is also proposed. To evaluate the accuracy of our proposed LF methods, the methods were compared with those of the nonlocal patch‐based LF method (Nonlocal‐PBM), the sparse patch‐based LF method (Sparse‐PBM), majority voting method, similarity and truth estimation for propagated segmentations, and hierarchical multi‐atlas LF with multi‐scale feature representation and label‐specific patch partition (HMAS). Based on our experimental results and quantitative comparison, our methods are promising in the magnetic resonance image segmentation.