Convolutional neural networks (CNNs) have become popular in medical image segmentation. The widely used deep CNNs are customized to extract multiple representative features for two-dimensional (2D) data, generally called 2D networks. However, 2D networks are inefficient in extracting three-dimensional (3D) spatial features from volumetric images. Although most 2D segmentation networks can be extended to 3D networks, the naively extended 3D methods are resource-intensive. In this paper, we propose an efficient and accurate network for fully automatic 3D segmentation. Specifically, we designed a 3D multiple-contextual extractor to capture rich global contextual dependencies from different feature levels. Then we leveraged an ROI-estimation strategy to crop the ROI bounding box. Meanwhile, we used a 3D ROI-attention module to improve the accuracy of in-region segmentation in the decoder path. Moreover, we used a hybrid Dice loss function to address the issues of class imbalance and blurry contour in medical images. By incorporating the above strategies, we realized a practical end-to-end 3D medical image segmentation with high efficiency and accuracy. To validate the 3D segmentation performance of our proposed method, we conducted extensive experiments on two datasets and demonstrated favorable results over the state-of-the-art methods.

This paper presents examination result of possibility for automatic unclear region detection in the CAD system for colorectal tumor with real time endoscopic video image. We confirmed that it is possible to realize the CAD system with navigation function of clear region which consists of unclear region detection by YOLO2 and classification by AlexNet and SVMs on customizable embedded DSP cores. Moreover, we confirmed the real time CAD system can be constructed by a low power ASIC using customizable embedded DSP cores.

In this paper, we present a classification method for a Computer-Aided Diagnosis (CAD) system in a colorectal magnified Narrow Band Imaging (NBI) endoscopy. In an endoscopic video image, color shift, blurring or reflection of light occurs in a lesion area, which affects the discrimination result by a computer. Therefore, in order to identify lesions with high robustness and stable classification to these images specific to video frame, we implement a CAD system for colorectal endoscopic images with the Convolutional Neural Network (CNN) feature and Support Vector Machine (SVM) classification on the embedded DSP core. To improve the robustness of CAD system, we construct the SVM learned by multiple image sizes data sets so as to adapt to the noise peculiar to the video image. We confirmed that the proposed method achieves higher robustness, stable, and high classification accuracy in the endoscopic video image. The proposed method also can cope with differences in resolution by old and new endoscopes and perform stably with respect to the input endoscopic video image.

A novel ROI-based reversible data hiding scheme is proposed for medical images, which is able to hide electronic patient record (EPR) and protect the region of interest (ROI) with tamper localization and recovery. The proposed scheme combines prediction error expansion with the sorting technique for embedding EPR into ROI, and the recovery information is embedded into the region of non-interest (RONI) using histogram shifting (HS) method which hardly leads to the overflow and underflow problems. The experimental results show that the proposed scheme not only can embed a large amount of information with low distortion, but also can localize and recover the tampered area inside ROI.

In this paper, an efficient multi-modal medical image fusion approach is proposed based on local features contrast and bilateral sharpness criterion in nonsubsampled contourlet transform (NSCT) domain. Compared with other multiscale decomposition analysis tools, the nonsubsampled contourlet transform not only can eliminate the “block-effect” and the “pseudo-effect”, but also can represent the source image in multiple direction and capture the geometric structure of source image in transform domain. These advantages of NSCT can, when used in fusion algorithm, help to attain more visual information in fused image and improve the fusion quality. At the same time, in order to improve the robustness of fusion algorithm and to improve the quality of the fused image, two selection rules should be considered. Firstly, a new bilateral sharpness criterion is proposed to select the lowpass coefficient, which exploits both strength and phase coherence. Secondly, a modified SML (sum modified Laplacian) is introduced into the local contrast measurements, which is suitable for human vision system and can extract more useful detailed information from source images. Experimental results demonstrate that the proposed method performs better than the conventional fusion algorithm in terms of both visual quality and objective evaluation criteria.

Surface integrity of 3D medical data is crucial for surgery simulation or virtual diagnoses. However, undesirable holes often exist due to external damage on bodies or accessibility limitation on scanners. To bridge the gap, hole-filling for medical imaging is a popular research topic in recent years [1]-[3]. Considering that a medical image, e.g. CT or MRI, has the natural form of a tensor, we recognize the problem of medical hole-filling as the extension of Principal Component Pursuit (PCP) problem from matrix case to tensor case. Since the new problem in the tensor case is much more difficult than the matrix case, an efficient algorithm for the extension is presented by relaxation technique. The most significant feature of our algorithm is that unlike traditional methods which follow a strictly local approach, our method fixes the hole by the global structure in the specific medical data. Another important difference from the previous algorithm [4] is that our algorithm is able to automatically separate the completed data from the hole in an implicit manner. Our experiments demonstrate that the proposed method can lead to satisfactory results.

Accurate registration is crucial for medical image analysis. In this letter, we proposed an improved Demons technique (IDT) for medical image registration. The IDT improves registration quality using orthogonal gradient information. The advantage of the proposed IDT is assessed using 14 medical image pairs. Experimental results show that the proposed technique provides about 8% improvement over existing Demons-based techniques in terms of registration accuracy.

Image registration is an important topic in medical image analysis. It is usually used in 2D mosaics to construct the whole image of a biological specimen or in 3D reconstruction to build up the structure of an examined specimen from a series of microscopic images. Nevertheless, owing to a variety of factors, including microscopic optics, mechanisms, sensors, and manipulation, there may be great differences between the acquired image slices even if they are adjacent. The common differences include the chromatic aberration as well as the geometry discrepancy that is caused by cuts, tears, folds, and deformation. They usually make the registration problem a difficult challenge to achieve. In this paper, we propose an efficient registration method, which consists of a feature-based registration approach based on analytic robust point matching (ARPM) and a refinement procedure of the feature-based Levenberg-Marquardt algorithm (FLM), to automatically reconstruct 3D vessels of the rat brains from a series of microscopic images. The registration algorithm could speedily evaluate the spatial correspondence and geometric transformation between two point sets with different sizes. In addition, to achieve subpixel accuracy, an FLM method is used to refine the registered results. Due to the nonlinear characteristic of FLM method, it converges much faster than most other methods. We evaluate the performance of proposed method by comparing it with well-known thin-plate spline robust point matching (TPS-RPM) algorithm. The results indicate that the ARPM algorithm together with the FLM method is not only a robust but efficient method in image registration.

We propose a modification of the generalized gradient vector flow field techniques based on multiresolution analysis and phase portrait techniques. The original image is subjected to mutliresolutional analysis to create a sequence of approximation and detail images. The approximations are converted into an edge map and subsequently into a gradient field subjected to the generalized gradient vector flow transformation. The procedure removes noise and extends large gradients. At every iteration the algorithm obtains a new, improved vector field being filtered using the phase portrait analysis. The phase portrait is applied to a window with a variable size to find possible boundary points and the noise. As opposed to previous phase portrait techniques based on binary rules our method generates a continuous adjustable score. The score is a function of the eigenvalues of the corresponding linearized system of ordinary differential equations. The salient feature of the method is continuity: when the score is high it is likely to be the noisy part of the image, but when the score is low it is likely to be the boundary of the object. The score is used by a filter applied to the original image. In the neighbourhood of the points with a high score the gray level is smoothed whereas at the boundary points the gray level is increased. Next, a new gradient field is generated and the result is incorporated into the iterative gradient vector flow iterations. This approach combined with multiresolutional analysis leads to robust segmentations with an impressive improvement of the accuracy. Our numerical experiments with synthetic and real medical ultrasound images show that the proposed technique outperforms the conventional gradient vector flow method even when the filters and the multiresolution are applied in the same fashion. Finally, we show that the proposed algorithm allows the initial contour to be much farther from the actual boundary than possible with the conventional methods.

We have developed a new computer-aided diagnostic system for differentiating oral borderline malignancies in hematoxylin-eosin stained microscopic images. Epithelial dysplasia and carcinoma in-situ (CIS) of oral mucosa are two different borderline grades similar to each other, and it is difficult to distinguish between them. A new image processing and analysis method has been applied to a variety of histopathological features and shows the possibility for differentiating the oral cancer borderline grades automatically. The method is based on comparing the drop-shape similarity level in a particular manually selected pair of neighboring rete ridges. It was found that the considered similarity level in dysplasia was higher than those in epithelial CIS, of which pathological diagnoses were conventionally made by pathologists. The developed image processing method showed a good promise for the computer-aided pathological assessment of oral borderline malignancy differentiation in clinical practice.

This paper proposes an improved method of reversible data hiding with increased capacity. The conventional method determines whether to embed a data bit in an image block according to the statistics of pixels in that block. Some images have pixel statistics that are inadequate for data hiding, and seldom or never have data embedded in them. The proposed method modulates the statistics invertibility to overcome such disadvantages, and is also able to improve the quality of the image containing the hidden data using block-adaptive modulation. Simulationresults show the effectiveness of the proposed method.

Bone Mineral Density (BMD) is an indicator of osteoporosis that is an increasingly serious disease, particularly for the elderly. To calculate BMD, we need to measure the volume of the femur in a noninvasive way. In this paper, we propose a noninvasive bone volume measurement method using x-ray attenuation on radiography and medical knowledge. The absolute thickness at one reference pixel and the relative thickness at all pixels of the bone in the x-ray image are used to calculate the volume and the BMD. First, the absolute bone thickness of one particular pixel is estimated by the known geometric shape of a specific bone part as medical knowledge. The relative bone thicknesses of all pixels are then calculated by x-ray attenuation of each pixel. Finally, given the absolute bone thickness of the reference pixel, the absolute bone thickness of all pixels is mapped. To evaluate the performance of the proposed method, experiments on 300 subjects were performed. We found that the method provides good estimations of real BMD values of femur bone. Estimates shows a high linear correlation of 0.96 between the volume Bone Mineral Density (vBMD) of CT-SCAN and computed vBMD (all P<0.001). The BMD results reveal 3.23% difference in volume from the BMD of CT-SCAN.

We present a watermarking scheme that combines data compression and encryption in application to radiological medical images. In this approach we combine the image moment theory and image homogeneity in order to recover the watermark after a geometrical distortion. Image quality is measured with metrics used in image processing, such as PSNR and MSE.

Image Registration can be seen as an optimization problem to find a cost function and then use an optimization method to get its minimum. Normalized mutual information is a widely-used robust method to design a cost function in medical image registration. Its calculation is based on the joint histogram of the fixed and transformed moving images. Usually, only a discrete joint histogram is considered in the calculation of normalized mutual information. The discrete joint histogram does not allow the cost function to be explicitly differentiated, so it can only use non-gradient based optimization methods, such as Powell's method, to seek the minimum. In this paper, a parzen-window based method is proposed to estimate the continuous joint histogram in order to make it possible to derive the close form solution for the derivative of the cost function. With this help, we successfully apply the gradient-based optimization method in registration. We also design a new kernel for the parzen-window based method. Our designed kernel is a second order polynomial kernel with the width of two. Because of good theoretical characteristics, this kernel works better than other kernels, such as a cubic B-spline kernel and a first order B-spline kernel, which are widely used in the parzen-window based estimation. Both rigid and non-rigid registration experiments are done to show improved behavior of our designed kernel. Additionally, the proposed method is successfully applied to a clinical CT-MR non-rigid registration which is able to assist a magnetic resonance (MR) guided microwave thermocoagulation of liver tumors.

This paper presents the design and implementation of a hip range of motion (ROM) estimation method that is capable of fine-grained estimation during total hip replacement (THR) surgery. Our method is based on two acceleration strategies: (1) adaptive mesh refinement (AMR) for complexity reduction and (2) parallelization for further acceleration. On the assumption that the hip ROM is a single closed region, the AMR strategy reduces the complexity for N N N stance configurations from O(N3) to O(ND), where 2≤D≤3 and D is a data-dependent value that can be approximated by 2 in most cases. The parallelization strategy employs the master-worker paradigm with multiple task queues, reducing synchronization between processors with load balancing. The experimental results indicate that the implementation on a cluster of 64 PCs completes estimation of 360360180 stance configurations in 20 seconds, playing a key role in selecting and aligning the optimal combination of artificial joint components during THR surgery.

The objective of this paper is to present a decision support system which uses a computer-based procedure to detect tumor blocks or lesions in digitized medical images. The authors developed a simple method with a low computation effort to detect tumors on T2-weighted Magnetic Resonance Imaging (MRI) brain images, focusing on the connection between the spatial pixel value and tumor properties from four different perspectives: 1) cases having minuscule differences between two images using a fixed block-based method, 2) tumor shape and size using the edge and binary images, 3) tumor properties based on texture values using spatial pixel intensity distribution controlled by a global discriminate value, and 4) the occurrence of content-specific tumor pixel for threshold images. Measurements of the following medical datasets were performed: 1) different time interval images, and 2) different brain disease images on single and multiple slice images. Experimental results have revealed that our proposed technique incurred an overall error smaller than those in other proposed methods. In particular, the proposed method allowed decrements of false alarm and missed alarm errors, which demonstrate the effectiveness of our proposed technique. In this paper, we also present a prototype system, known as PCB, to evaluate the performance of the proposed methods by actual experiments, comparing the detection accuracy and system performance.

In this paper, we propose a construction method of three-dimensional deformable models that represent tree-shaped human organs, such as bronchial tubes, based on results obtained by statistically analyzing the distributions of bifurcation points in the tree-shaped organs. The models are made to be used as standard templates of tree-shaped organs in medical image recognition, and are formed by control points that can be uniquely identified as structural elements of organs such as bifurcation tracheae in bronchial tubes. They can be transfigured based on the statistical validity of relationships between the control points. The optimal state of that transfiguration is determined within the framework of energy minimization. Experimental results from bronchial tubes are shown on actual CT images.

Malignant melanoma is a skin cancer that can be mistaken as a harmless mole in the early stages and is curable only in these early stages. Therefore, dermatologists use a microscope that shows the pigment structures of the skin to classify suspicious skin lesions as malignant or benign. This microscope is called "dermoscope." However, even when using a dermoscope a malignant skin lesion can be mistaken as benign or vice versa. Therefore, it seems desirable to analyze dermoscopic images by computer to classify the skin lesion. Before a dermoscopic image can be classified, it should be segmented into regions of the same color. For this purpose, we propose a segmentation method that automatically determines the number of colors by optimizing a cluster validity index. Cluster validity indices can be used to determine how accurately a partition represents the "natural" clusters of a data set. Therefore, cluster validity indices can also be applied to evaluate how accurately a color image is segmented. First the RGB image is transformed into the L*u*v* color space, in which Euclidean vector distances correspond to differences of visible colors. The pixels of the L*u*v* image are used to train a self-organizing map. After completion of the training a genetic algorithm groups the neurons of the self-organizing map into clusters using fuzzy c-means. The genetic algorithm searches for a partition that optimizes a fuzzy cluster validity index. The image is segmented by assigning each pixel of the L*u*v* image to the nearest neighbor among the cluster centers found by the genetic algorithm. A set of dermoscopic images is segmented using the method proposed in this research and the images are classified based on color statistics and textural features. The results indicate that the method proposed in this research is effective for the segmentation of dermoscopic images.

In this paper, we propose a new computer-aided diagnosis system which can extract specific features from hematoxylin and eosin (HE)-stained breast tumor images and evaluate the type of tumor using artificial organisms. The gene of the artificial organisms is defined by three kinds of texture features, which can evaluate the specific features of the tumor region in the image. The artificial organisms move around in the image and investigate their environmental conditions during the searching process. When the target pixel is regarded as a tumor region, the organism obtains energy and produces offspring; organisms in other regions lose energy and die. The searching process is iterated until the 30th generation; as a result, tumor regions are filled with artificial organisms. Whether the detected tumor is benign or malignant is evaluated based on the combination of selected genes. The method developed was applied to 27 test cases and the distinction between benign and malignant tumors by the artificial organisms was successful in about 90% of tumor images. In this diagnosis support system, the combination of genes, which represents specific features of detected tumor region, is selected automatically for each tumor image during the searching process.

We propose new lossless medical image compression method based on hierarchical sorting technique. Hierarchical sorting is a technique to achieve high compression ratio by detecting the regions where image pattern varies abruptly and sorting pixel order by its value to increase predictability. In this method, we can control sorting accuracy along with size and complexity. As the result, we can reduce the sizes of the permutation-tables and reuse the tables to other image regions. Comparison using experimental implementation of this method shows better performance for medical image set measured by X-ray CT and MRI instruments where similar sub-block patterns appear frequently. This technique applies quad-tree division method to divide an image to blocks in order to support progressive decoding and fast preview of large images.