E-Liang Chen

An automatic diagnostic system for CT liver image classification

Computed tomography (CT) images have been widely used for liver disease diagnosis. Designing and developing computer-assisted image processing techniques to help doctors improve their diagnosis has received considerable interests over the past years. In this paper, a CT liver image diagnostic classification system is presented which will automatically find, extract the CT liver boundary and further classify liver diseases. The system comprises a detect-before-extract (DBE) system which automatically finds the liver boundary and a neural network liver classifier which uses specially designed feature descriptors to distinguish normal liver, two types of liver tumors, hepatoma and hemageoma. The DBE system applies the concept of the normalized fractional Brownian motion model to find an initial liver boundary and then uses a deformable contour model to precisely delineate the liver boundary. The neural network is included to classify liver tumors into hepatoma and hemageoma. It is implemented by a modified probabilistic neural network (PNN) [MPNN] in conjunction with feature descriptors which are generated by fractal feature information and the gray-level co-occurrence matrix. The proposed system was evaluated by 30 liver cases and shown to be efficient and very effective.

Polygonal approximation using a competitive Hopfield neural network

Abstract Polygonal approximation plays an important role in pattern recognition and computer vision. In this paper, a parallel method using a Competitive Hopfield Neural Network (CHNN) is proposed for polygonal approximation. Based on the CHNN, the polygonal approximation is regarded as a minimization of a criterion function which is defined as the arc-to-chord deviation between the curve and the polygon. The CHNN differs from the original Hopfield network in that a competitive winner-take-all mechanism is imposed. The winner-take-all mechanism adeptly precludes the necessity of determining the values for the weighting factors in the energy function in maintaining a feasible result. The proposed method is compared to several existing methods by the approximation error norms L2 and L∞ with the result that promising approximation polygons are obtained.

FALL DETECTION USING MODULAR NEURAL NETWORKS WITH BACK-PROJECTED OPTICAL FLOW

This paper presents a video-based algorithm for fall detection used the modular neural networks with the motion vectors computed by block-based optical flow back-projection (BOFB). From a video sequence, the moving object is extracted first and the pixels with high intensity variance in the extracted object are determined as feature points. The motion vector in this application is required to represent the actual motion displacement, rather than visually significant similarity. Therefore, we proposed BOFB which back-projects optical flows in a block to restore the motion vector from gradient-based optical flows that is employed to estimate the genuine motion of these feature points. The sequences of feature vectors are fed into the proposed Time-Delay Hierarchical modular Neural Network (TDHNN) for fall detection. The TDHNN consists of two major modular networks: several Time-Delay Neural Networks (TDNNs) trained by various feature characteristics and a Support Vector Machine (SVM) for final decision. This paper also purposed Slide Window Accumulate (SWA) mechanism for the increase of the robustness of the system in fall detection. The experimental results show that the proposed algorithm is efficacious and reliable in fall detection.

A region-based selective optical flow back-projection for genuine motion vector estimation

Motion vector plays one significant feature in MPEG-4 video object segmentation. However, the motion vector in this application is required to represent the actual motion displacement, rather than regions of visually significant similarity. Optical flow back-projection (OFB), which back-projects optical flows in a cluster to restore the regions motion vector from gradient-based optical flows, is proposed to obtain genuine motion displacement. The back-projection is performed based on minimizing the projection mean square errors of the motion vector on gradient directions. As optical flows of various magnitudes and directions provide various degrees of reliability in the genuine motion restoration, the optical flows to be used in the OFB are selected based on high gradient. With this approach, the interference by similar blocks, which frequently results in irrational motion displacement in the traditional block matching approach, can be avoided and more reliable motion vectors can be obtained

A spatiotemporal neural network for recognizing partially occluded objects

In this paper, a spatiotemporal neural network for partially occluded object recognition is presented. The system consists of two major components: a feature extraction process and a spatiotemporal modular neural network. The former is made up of a sequence of preprocessing techniques including thresholding, boundary extraction, Gaussian filtering, and a split-and-merge algorithm to generate features that will represent the objects to be recognized. These acquired features are invariant to rotation, translation, and scaling and can serve as input to the spatiotemporal network that utilizes the concept of tap delay to account for spatial correlation between consecutive input features. A shape perceiver is designed into the network to extract continued parts of an object as well as to enable the inclusion of each objects unique characteristics into the system. Traditional neural network approaches for recognizing partially occluded objects have encountered significant problems because of the incomplete boundaries of the objects. In our approach, by creatively installing tap delays, the system can escape this limitation. Experimental results show that the proposed system can produce satisfactory results in efficiently and effectively recognizing partially occluded objects. Furthermore, intrinsic to this system is the ease by which it can be realized through parallel implementation, thus minimizing the tremendous time spent in matching object contours stored in a model database, as is the case in conventional recognition systems.

Using a fuzzy engine and complete set of features for hepatic diseases diagnosis: Integrating contrast and non-contrast CT images

In the diagnosis of hepatic diseases, “Contrast-Enhanced Computerized Tomography” (CECT) and “Non-Contrast CT” (NCT) are usually simultaneously adopted. In this paper, a system consisting of a fuzzy diagnosis engine and a feature quantizer, which extracts hepatic features from CECT and NCT images, is proposed for assisting hepatic disease diagnosis. Compared with existing methods, this paper differs in two folds. First, a more complete feature set composed of not only lesion textures, but also lesion morphological structure and lesion contrast to normal tissues is used. These features are described through mathematical models built inside the feature quantizer and served as the input of fuzzy diagnosis engine. Second, because of the use of the fuzzy diagnosis engine, the system is intrinsically with the capability of storing rules and may infer and adapt its rules according to learning data. Furthermore, uncertainty associated with disease diagnosis can be appropriately taken into considerations. The system has been tested using 131 sets of image data, which are to be classified into 4 types of diseases: liver cyst, hepatoma, cavernous hemagioma and metastatic liver tumor. Experimental results indicate that among these test data 78% of them are accurately classified as one type, while the remaining 22% of data are classified as more than one types of diseases. Even so, within these 22% of multiple-classified data, the correct type is always included in the output in each test, showing a promise of the system.

Optical flow back-projection for genuine motion vector estimation

Motion vector plays one significant feature in MPEG-4 video object segmentation. However, the motion vector in this application is required to represent the actual motion displacement, rather than regions of visually significant similarity. Optical flow back-projection (OFB), which back-projects optical flows in a cluster to restore the regions motion vector from gradient-based optical flows, is proposed to obtain genuine motion displacement. The back-projection is performed based on minimizing the projection mean square errors of the motion vector on gradient directions. As optical flows of various magnitudes and directions provide various degrees of reliability in the genuine motion restoration, the optical flows to be used in the OFB are selected based on high gradient. With this approach, the interference by similar blocks, which frequently results in irrational motion displacement in the traditional block matching approach, can be avoided and more reliable motion vectors can be obtained

Responses of central auditory neurons modeled with finite impulse response (FIR) neural networks

To simulate central auditory responses to complex sounds, a computational model was implemented. It consists of a multi-scale classification process, and an artificial neural network composed of two modules of finite impulse response (FIR) neural networks connected to a maximum network. Electrical activities of single auditory neurons were recorded at the rat midbrain in response to a repetitive pseudo-random frequency modulated (FM) sound. The multi-scale classification process divides the training dataset into either strong or weak response using a multiple-scale Gaussian filter that based on response probability. Two modules of FIR neural network are then independently trained to model the two types of responses. This caters for the possible differences in neuronal circuitry and transmission delay. Their outputs are connected to a maximum network to generate the final output. After training, we use a different set of FM responses collected from the same neuron to test the performance of the model. Two criteria are adopted for assessment. One measures the matching of the modeled output to the actual output on a point-to-point basis. Another measures the matching of bulk responses between the two. Results show that the proposed model predicts the responses of central auditory neurons satisfactorily.

Pattern recognition using a hierarchical neural network

A pattern recognition system based on hierarchical neural networks is proposed in this paper. The hierarchical system consists of two levels of networks: low-level for feature extraction and high-level for object recognition. The low-level is a competitive Hopfield neural network (CHNN) which detects the dominant points of a target shape to be the pattern features, based on the minimization of a cost function. The CHNN is implemented by incorporating a winner-take-all strategy in the network. By imposing the winner-take-all rule, one is relieved from deciding the suitable values of the weighting factors in the cost function. Furthermore, from the experimental results, the authors also find that the proposed CHNN performs very well in determining the dominant points of a target shape. After the features have been extracted, they are applied to the high-level multilayered network for object recognition. Because the multilayer network has high robustness to the pattern variations, the recognition system is found to possess high noise tolerance capability. Experimental results show that the system can recognize all the objects correctly when the percentage of noises is under 10%. Even when the percentage of noises reaches 40%, the recognition ratio is still over 90%.<<ETX>>

Detecting Sustained Attention during Cognitive Work Using Heart Rate Variability

Sustained attention is an important requirement when we are doing vigilant works. The detection of whether a worker is in sustained attention stage is important to maintain the safety of the worker. Thus this paper performs classification of sustained attention and non sustained attention phases based on the heart rate variability (HRV). To achieve this purpose several features are derived from time domain, frequency domain and nonlinear analysis from Electrocardiogram (ECG). Then linear discriminant analysis (LDA) with Knearest neighbor (KNN) is adopted as the classifier. It was found that the proposed method is promising in classifying the sustained attention and non sustained attention with 98% of accuracy.