D.D. Feng

Automatic 3D temporal kinetics segmentation of dynamic emission tomography image using adaptive region growing cluster analysis

The primary goal of medical image segmentation is to partition the raw image into region of interests (ROIs) matching the anatomical localization of objects of interest in 2D or 3D space. The traditional method of ROI delineation (or segmentation) for the analysis of dynamic emission tomography is the manual placement of ROIs by the operator. However this approach is operator dependent, time-consuming and may lack good reproducibility. Quantitative positron emission tomography (PET) studies can provide measurements of dynamic physiological and biochemical processes in humans through the use of temporal kinetics available. However, due to the relatively poor spatial resolution and high noise levels, partitioning of ROIs is limited. In this paper, the use of a novel knowledge-based approach to segmentation of clinical PET studies using automatic seed selection for adaptive region growing based on Euclidean distance between the local tissue time-activity curves (TTAC) of the voxels is proposed.

Medical Image Registration via Steerable Pyramid

Image registration is enabling in integrating complementary and heterogeneous information from multiple images, and is particularly important for high-quality healthcare. To improve registration efficiency and accuracy, in this paper, a two-resolution-scale registration approach is proposed. Firstly, to speed up calculation, the images will be decomposed into multi-scale and multi-band representation by steerable pyramid that outweighs wavelets by providing invariance for both translation and rotation. Then, to avoid transformation error accumulation and magnification during the parameter transmission in the traditional multi-scale registration, the registration will be performed only in the lowest-resolution scale and the highest-resolution scale. In the former scale, the global rotation and scaling parameters will be calculated rapidly and accurately, which then will be directly used to initialize optimization in the latter scale, where, the translation differences will be corrected. The experiments on medical images demonstrate that the proposed registration is of good performance.

Multiple block based authentication watermarking for distribution of medical images

To provide an effective authentication method for distribution of medical images, we propose a new fragile watermarking method utilizing segmentation information of medical image contents. For medical image modalities like CT, MRI and PET, tissue structures contain a significant amount of clinical information. Hence, it is important to provide an authenticity check of the segmented blocks. The proposed method is based on cryptographically secure fragile watermarking but eliminates the problem of the block-wise independency of existing methods. The vector quantization (VQ) counterfeiting attack is a known vulnerability due to the block-wise independency of existing methods. In our approach, multiple signatures from two different types of blocks are used to defy such attacks. More secure distribution of medical images can be achieved by embedding the watermark.

Volume of interest (VOI) feature representation and retrieval of multi-dimensional dynamic positron emission tomography images

Medical images often require domain knowledge and diagnostic features in order to measure the similarity and relevance. It is unlikely that conventional image feature representation can adequately represent the volumetric and diagnostic properties of the medical images for content-based retrieval. In this study, we present a volume of interest (VOI) feature representation and retrieval of multi-dimensional dynamic positron emission tomography (PET) images consisting of 3D spatial and 1D temporal domains. The VOIs are represented by location based on the 3D standard atlas, physiological temporal kinetics and parametric estimate, volume, and surface distribution by utilizing both the spatial and temporal domain, in addition to the textual information. The proposed representation of the features can allow for effective retrieval by domain specific image features and are applicable for content-based image retrieval (CBIR).

The accurate extraction and tracking of moving objects for video surveillance

In this paper, an accurate extraction approach and an effective tracking algorithm for moving objects are proposed. We first describe our extraction algorithm, which can extract objects precisely by using the mixed differences between adaptive background estimation and the two adjacent frames. Then we design a tracking algorithm, in which we use several IIR filters to predict the moving objects position and improve efficiency of the objects matching by using the moving objects temporal image and extracting results. The proposed tracking algorithm can track the objects accurately even if the objects are in the process of acceleration and have slight deformations. The efficiency of the proposed algorithms is demonstrated by several experiments given in this paper.

Dual-stage deep learning framework for pigment epithelium detachment segmentation in polypoidal choroidal vasculopathy

Worldwide, polypoidal choroidal vasculopathy (PCV) is a common vision-threatening exudative maculopathy, and pigment epithelium detachment (PED) is an important clinical characteristic. Thus, precise and efficient PED segmentation is necessary for PCV clinical diagnosis and treatment. We propose a dual-stage learning framework via deep neural networks (DNN) for automated PED segmentation in PCV patients to avoid issues associated with manual PED segmentation (subjectivity, manual segmentation errors, and high time consumption).The optical coherence tomography scans of fifty patients were quantitatively evaluated with different algorithms and clinicians. Dual-stage DNN outperformed existing PED segmentation methods for all segmentation accuracy parameters, including true positive volume fraction (85.74 ± 8.69%), dice similarity coefficient (85.69 ± 8.08%), positive predictive value (86.02 ± 8.99%) and false positive volume fraction (0.38 ± 0.18%). Dual-stage DNN achieves accurate PED quantitative information, works with multiple types of PEDs and agrees well with manual delineation, suggesting that it is a potential automated assistant for PCV management.

Enhancing home area networks to facilitate telehealth services: Test-bed scenario with video-consultation-calls (VCC)

Recent years have witnessed rapid development of Telehealth applications and the integration of portable devices to facilitate services such as Voice-over-IP (VoIP), Video-over-IP and distribution of other data types e.g., medical images and health sensor data. In line with Telehelath, Home Area Network (HAN) has also seen broad deployments and has expanded conventional households to support latest electronic devices that are equipped with network connectivity. Telehealth applications are starting to be designed to take advantages of the HAN environment and are providing healthcare at the patients home, i.e., facilitates hospital at home via video-consultation-call (VCC). However, some of these Telehealth applications are highly delay sensitive and require higher bandwidth priority and a network load-balancing system in order to produce best quality of service (QoS). We observed that typical HAN is established without bandwidth priority and network-load-balancing systems. This lack of management causes HAN to produce unacceptable QoS for Telehealth applications. In this scenario, we propose a scheme to enhance HAN while providing better quality of VCC to support Telehealth services. Our scheme controls the Internet traffic within the HAN network infrastructure environment. In our experimental setting, we implemented a real network test-bed based on HAN where two-way Skype VCC was established and the quality of packet-loss was measured. Skype was chosen for its wide availability and support on all types of computing devices (PC, mobile, laptop etc.). Our results indicate that packet-loss fluctuates between ~5-16% when user datagram protocol (UDP) traffic is utilised concurrently by other devices in HAN. It also demonstrated that our scheme could enhance HAN performance for VCC as it reduced packet-loss down to ~0.3 %.

Face recognition using hybrid feature

In this paper, a face recognition algorithm using hybrid feature is proposed. The algorithm consists of three steps. In the first step, a coarse classification is performed using Mutual Information match. In the second step, the Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) features of frequency spectrum are extracted, which will be taken as the input of the RBFN in the third step. In the last step, the classification is carried out by using RBFN. The proposed approach has been tested on ORL face database and Shimon database. The experimental results have demonstrated that the performance of this algorithm is superior to the other algorithms on the same database.

Optimal design of scan intervals for glucose kinetic modeling using dynamic imaging data from PET

In tracer kinetic modeling with positron emission tomography (PET) it is required to take the tracer tissue time-activity images which are of accumulated counts corresponding to the scan intervals. In this paper, the optimal design of the scan intervals to estimate metabolic rate of glucose (MRGlc) is investigated. Specific features of the PET measurements are considered. The results show that, if the scan intervals are selected properly, the number of dynamic images required can be reduced by approximately 75%, the time for data analysis can be reduced, and the accuracy of parameter estimation can be improved.<<ETX>>

Hierarchical elastic registration of human brain images based on wavelet decomposition

Wavelets have been studied and applied in the multimedia and video processing community, including some areas in medical imaging, however, wavelet-based medical image registration has not been explored properly. In this paper, we propose an hierarchical registration method based on wavelet decomposition. Firstly, based on the wavelets, the images are decomposed into subbands which compose the registration pyramid; then, in each hierarchy, the affine registration based on mutual information is performed and the results of the current registration are used as the initial guess for the next hierarchy; finally, to further improve the registration performance, the local elastic registration is carried out. The proposed algorithm has been validated by experiments on clinical tomographic images and our experimental results demonstrate the the proposed method is of efficiency and good accuracy.