Guoyuan Liang

Volumetric Topological Analysis: A Novel Approach for Trabecular Bone Classification on the Continuum Between Plates and Rods

Trabecular bone (TB) is a complex quasi-random network of interconnected plates and rods. TB constantly remodels to adapt to the stresses to which it is subjected (Wolffs Law). In osteoporosis, this dynamic equilibrium between bone formation and resorption is perturbed, leading to bone loss and structural deterioration. Both bone loss and structural deterioration increase fracture risk. Bones mechanical behavior can only be partially explained by variations in bone mineral density, which led to the notion of bone structural quality. Previously, we developed digital topological analysis (DTA) which classifies plates, rods, profiles, edges, and junctions in a TB skeletal representation. Although the method has become quite popular, a major limitation of DTA is that it provides only hard classifications of different topological entities, failing to distinguish between narrow and wide plates. Here, we present a new method called volumetric topological analysis (VTA) for regional quantification of TB topology. At each TB location, the method uniquely classifies its topology on the continuum between perfect plates and perfect rods, facilitating early detections of TB alterations from plates to rods according to the known etiology of osteoporotic bone loss. Several new ideas, including manifold distance transform, manifold scale, and feature propagation have been introduced here and combined with existing DTA and distance transform methods, leading to the new VTA technology. This method has been applied to multidetector computed tomography (CT) and micro-computed tomography (μCT) images of four cadaveric distal tibia and five distal radius specimens. Both intra- and inter-modality reproducibility of the method has been examined using repeat CT and μCT scans of distal tibia specimens. Also, the methods ability to predict experimental biomechanical properties of TB via CT imaging under in vivo conditions has been quantitatively examined and the results found are very encouraging.

A New Osteophyte Segmentation Algorithm Using the Partial Shape Model and Its Applications to Rabbit Femur Anterior Cruciate Ligament Transection via Micro-CT Imaging

Osteophyte is an additional bony growth on a normal bone surface limiting or stopping motion at a deteriorating joint. Detection and quantification of osteophytes from computed tomography (CT) images is helpful in assessing disease status as well as treatment and surgery planning. However, it is difficult to distinguish between osteophytes and healthy bones using simple thresholding or edge/texture features due to the similarity of their material composition. In this paper, we present a new method primarily based on the active shape model (ASM) to solve this problem and evaluate its application to the anterior cruciate ligament transaction (ACLT) rabbit femur model via micro-CT imaging. The common idea behind most ASM-based segmentation methods is to first build a parametric shape model from a training dataset and then apply the model to find a shape instance in a target image. A common challenge with such approaches is that a diseased bone shape is significantly altered at regions with osteophyte deposition misguiding an ASM method and eventually leading to suboptimum segmentations. This difficulty is overcome using a new partial-ASM method that uses bone shape over healthy regions and extrapolates it over the diseased region according to the underlying shape model. Finally, osteophytes are segmented by subtracting partial-ASM-derived shape from the overall diseased shape. Also, a new semiautomatic method is presented in this paper for efficiently building a 3-D shape model for an anatomic region using manual reference of a few anatomically defined fiducial landmarks that are highly reproducible on individuals. Accuracy of the method has been examined on simulated phantoms while reproducibility and sensitivity have been evaluated on micro-CT images of 2-, 4- and 8-week post-ACLT and sham-treated rabbit femurs. Experimental results have shown that the method is highly accurate (R\bm 2=0.99), reproducible (ICC = 0.97), and sensitive in detecting disease progression (p values: 0.065, 0.001, and <;0.001 for 2 weeks versus 4 weeks, 4 weeks versus 8 weeks, and 2 weeks versus 8 weeks, respectively).

Reproducibility of subregional trabecular bone micro-architectural measures derived from 7-Tesla magnetic resonance images

High-resolution magnetic resonance imaging (MRI) of trabecular bone combined with quantitative image analysis represents a powerful technique to gain insight into trabecular bone micro-architectural derangements in osteoporosis and osteoarthritis. The increased signal-to-noise ratio of ultra high-field MR (≥7 Tesla) permits images to be obtained with higher resolution and/or decreased scan time compared to scanning at 1.5/3T. In this small feasibility study, we show high measurement precision for subregional trabecular bone micro-architectural analysis performed on 7T knee MR images. The results provide further support for the use of trabecular bone measures as biomarkers in clinical studies of bone disorders.

A 3D object recognition and pose estimation system using deep learning method

This paper addresses a 3D object recognition and pose estimation method with a deep learning model. We train two separated Deep Belief Networks (DBN) before connecting the last layers together to train a classifier. By this means, we can simplify the complicated 3D problem to an easier classifier training problem. The deep learning model shows its advantages in learning hierarchical features which greatly facilitate the recognition mission. We apply the new Deep Belief Networks that combine the two traditional DBNs together and assign different poses of objects as different classes in the system. Besides, to overcome the shortcoming in object detection of the deep learning model, a new object detection method based on K-means clustering is presented. We have built a database comprised of 4 objects with different poses and illuminations for experimental performance evaluation. The experimental results demonstrate that our system with two cameras using the new DBNs can achieve high accuracy on 3D object recognition as well as pose estimation.

Quantification of stenosis in coronary artery via CTA using fuzzy distance transform

tomographic angiography (CTA) being noninvasive, economical and informative, has become a common modality for monitoring disease status and treatment effects. Here, we present a new method for detecting and quantifying coronary arterial stenosis via CTA using fuzzy distance transform (FDT) approach. FDT computes local depth at each image point in the presence of partial voluming. Coronary arterial stenoses are detected and their severities are quantified by analyzing FDT values along the medial axis of an artery obtained by skeletonization. Also, we have developed a new skeletal pruning algorithm toward improving quality of medial axes and therefore, enhancing the accuracy of stenosis detection and quantification. The method is completed using the following steps - (1) fuzzy segmentation of coronary artery via CTA, (2) FDT computation of coronary arteries, (3) medial axis computation, (4) estimation of local diameter along arteries and (5) stenosis detection and quantification of arterial blockage. Performance of the method has been quantitatively evaluated on a realistic coronary artery phantom dataset with randomly simulated stenoses and the results are compared with a classical binary algorithm. The method has also been applied on a clinical CTA dataset from thirteen patients with 59 stenoses and the results are compared with an experts quantitative assessment of stenoses. Results of the phantom experiment indicate that the new method is significantly more accurate as compared to the conventional binary method. Also, the results of the clinical study indicate that the computerized method is highly in agreement with the experts assessments.

A real-time dynamic hand gesture recognition system using kinect sensor

The use of hand gestures provides an attractive alternative to cumbersome interface devices for Human-Computer Interaction (HCI). However, in dynamic gesture recognition area, hand tracking under a complicated environment and gesture spotting namely detecting the start and end point are the two most challenging topics. In our work, a realtime Kinect-based dynamic hand gesture recognition (HGR) system which contains hand tracking, data processing, model training and gesture classification is proposed. In the first stage, two states of the performed hand including open and closed are utilized to achieve gesture spotting and 3D motion trajectories of gestures are captured by Kinect sensor. Further, motion orientation is extracted as the unique feature and Support Vector Machine (SVM) is used as the recognition algorithm in the proposed system. The results of experiments conducted in our database containing 10 Arabic numbers from 0 to 9 and the 26 characters of alphabet show efficiency with an average recognition rate of 95.42% and real-time performance of our method.

Infrared video based non-invasive heart rate measurement

In this paper, a non-invasive heart rate detection system with Kinect is developed. Kinect is used to measure the heart rate via the obtained near-infrared video. The color of the facial skin will fluctuate due to blood circulation, which can be extracted through image processing and empirical mode decomposition technology. The information included heart rate can thus be obtained. Compared with traditional detection methods and existing non-contact detection methods, the proposed one can measure the heart rate with high accuracy, convenience and robustness, especially in the complex illumination environments. Experiments in different scenarios have been performed to verify the efficacy of the proposed algorithm.

Anomaly detection and localization in crowded scenes using short-term trajectories

In this paper we present a method to detect and localize abnormal events in crowded scene. Most existing methods use the patch of optical flow or human tracking based trajectory as representation for crowd motion, which inevitably suffer from noises. Instead, we propose the employment of a new and efficient feature, short-term trajectory, which represent the motion of the visible and constant part of human body that move consistently, for modeling the complicated crowded scene. To extract the short-term trajectory, 3D mean-shift is firstly used to smooth the video frames and 3D seed filling algorithm is performed. In order to detect the abnormal events, all short-term trajectories are treated as point set and mapped into the image plane to obtain probability distribution of normalcy for every pixel. A cumulative energy is calculated based on these probability distributions to identify and localize the abnormal event. Experiments are conducted on known crowd data sets, and the results show that our method can achieve high accuracy in anomaly detection as well as effectiveness in anomalies localization.

A new multi‐object image thresholding method based on correlation between object class uncertainty and intensity gradient

PURPOSE Image thresholding and gradient analysis have remained popular image preprocessing tools for several decades due to the simplicity and straight-forwardness of their definitions. Also, optimum selection of threshold and gradient strength values are hidden steps in many advanced medical imaging algorithms. A reliable method for threshold optimization may be a crucial step toward automation of several medical image based applications. Most automatic thresholding and gradient selection methods reported in literature primarily focus on image histograms ignoring a significant amount of information embedded in the spatial distribution of intensity values forming visible features in an image. Here, we present a new method that simultaneously optimizes both threshold and gradient values for different object interfaces in an image that is based on unification of information from both the histogram and spatial image features; also, the method works for unknown number of object regions. METHODS A new energy function is formulated by combining the object class uncertainty measure, a histogram-based feature, of each pixel with its image gradient measure, a spatial contextual feature in an image. The energy function is designed to measure the overall compliance of the theoretical premise that, in a probabilistic sense, image intensities with high class uncertainty are associated with high image gradients. Finally, it is expressed as a function of threshold and gradient parameters and optimum combinations of these parameters are sought by locating pits and valleys on the energy surface. A major strength of the algorithm lies in the fact that it does not require the number of object regions in an image to be predefined. RESULTS The method has been applied on several medical image datasets and it has successfully determined both threshold and gradient parameters for different object interfaces even when some of the thresholds are almost impossible to locate in the histogram. Both accuracy and reproducibility of the method have been examined on several medical image datasets including repeat scan 3D multidetector computed tomography (CT) images of cadaveric ankles specimens. Also, the new method has been qualitatively and quantitatively compared with Otsus method along with three other algorithms based on minimum error thresholding, maximum segmented image information and minimization of homogeneity- and uncertainty-based energy and the results have demonstrated superiority of the new method. CONCLUSIONS We have developed a new automatic threshold and gradient strength selection algorithm by combining class uncertainty and spatial image gradient features. The performance of the method has been examined in terms of accuracy and reproducibility and the results found are better as compared to several popular automatic threshold selection methods.

Anomaly detection in crowds using a space MRF with incremental updates

In this paper, we propose a space Markov Random Field (MRF) model to detect abnormal activities in crowded scenes. The nodes of MRF graph consist of monitors evenly spread on the image, and neighboring nodes in space are associated with links. The normal patterns of activity at each node are learnt by constructing a Gaussian Mixture Model (GMM) upon optical flow locally, while correlation between adjacent nodes is represented by building a single Gaussian model upon inner product of histogram vectors of optical flow observed from a region centered at each node respectively. For any optical flow patterns detected in test video clips, we use the learnt model and MRF graph to calculate an energy value for each local node, and determine whether the behavior pattern of the node is normal or abnormal by comparing the value with a threshold. Further, we apply a method similar to updating of GMM for background subtraction to incrementally update the current model to adapt for visual context changes over a long period of time. Experiments on the published UCSD anomaly datasets Ped1 and Ped2 show the effectiveness of our method.