Abstract:A feature selection algorithm using Quadratic Programming is proposed in this paper based on feature margins. First, the inner-class distance of features is taken as the coefficients of the quadratic terms in the objective function and the inter-class distance of features is used as the coefficients of the linear terms for searching informative features. The elements of the quadratic terms and the linear terms are normalized to balance the feature relation between inner class and inter-class. Then, the optimal solution vector is taken as the feature weight vector for selecting informative features. Experiments on 6 different datasets showed the effectiveness and feasibility of the proposed method.

Abstract:In order to eliminate the ECG interference included in diaphragmatic electromyographic (EMGdi), an "inverse"hard threshold cancellation algorithm based on stationary wavelet transform was presented in this paper. Firstly,QRS detection and ECG interference range were made by wavelet coefficients after decomposing clinical EMGdi. Secondly, the mean square was estimated by wavelet coefficients in the ECG cycle at each level, and each ECG interference threshold was calculated by mean square. Finally, ECG interference wavelet coefficients were removed by "inverse" hard threshold, and then the denoised signal were reconstructed by wavelet coefficients. The experimental results show that the method cancels the ECG interference in EMGdi availably and reserve its signal characteristics effectively.

Abstract:Ultrasound liver images have characteristics such as wide dynamic range, abundant details and low contrast, which are not propitious for features extraction and pattern classification. Images enhancement algorithms based on Multi-Scale Retinex (MSR) combines the dynamic range compression with the tonal rendition at the same time. We apply three-scale Retinex model to enhance ultrasonic liver images, then extract the lightness, contrast, and entropy of the enhanced images, comparing the results with that of the usual image enhancement methods, such as histogram equalization algorithm and homomorphic filtering algorithm. Experiment results show that image enhancement algorithms based on MSR model can improve the distinguish degree of ultrasonic liver images features (contrast, entropy), and increase contrast and entropy, which is helpful to the assisted diagnosis of liver diseases.

Abstract:Laser Scanning Confocal Microscope(LSCM)，which is also called ‘Cell CT’, can be used to acquire image stacks of live cancer cells. These images are of great significance to fulfill the 3D-visualization on cell level. In this study, we proposed a 3D interpolation method based on contour points in polar coordinate system. In the new method, the origin of coordinate system is chosen to be the morphological center instead of the traditional geometric center. Resample the contour points by polar angle first, then fit the discrete points with uniform cubic B-spline to perform the interpolation, which will integrate multi-slice counter shape information. For a common nucleus structure, branched nucleus, a special method combining image morphological methods and corner detection methods based on curvature scale space is considered to solve contour division issue, and their correspondence problem was solved through a threshold radius method. The 3D reconstruction result shows much smoother and natural surface of cells and nucleus. This interpolation method is good for precise measurement of cell morphology and clinic study as well.

Abstract:This paper presents a reversible or lossless watermarking algorithm base on parity for medical image. Compared to the general image medical image have more close to the 0-value pixels which are easy to overflow in the embedding process. By embedding test, select pixels which are able to carry information to embed watermarking information. This algorithm takes full advantage of all pixels which are able to carry watermark information, so the embedded rate of the image is enhanced. To ensure the correct classification of pixels before extracting and the reversibility of the algorithm, the testing algorithm based on the parity of data is improved. Experiments show that this algorithm can both improve the watermarking capacity of the image and guarantee visual effect. For general images, both watermark embedding and visual effect have some improvement.

Abstract:TMS320C6455 is a high performance digital signal processor (DSP) provided by Texas Instruments Corporation of USA. By using the low-cost C6455 chip as the core processor, a filtered back projection (FBP) based positron emission tomography (PET) image reconstruction algorithm is optimized and implemented in this paper. By specific effective optimization of algorithm and DSP code, the system can reconstruct a 512×512 PET image in 40 seconds, and achieve desirable image definition. The PET image reconstruction method based on DSP can be applied to the area of PET medicine image reconstruction and other related areas.

Abstract:In the study of auditory evoked-potentials (AEPs) under high stimulus rate paradigm, the generation of appropriate stimulus sequences in an automatic way will facilitate the research and application of such paradigm. This paper presents a method to obtain the applicable stimulus-sequence using a modified differential evolution (DE) algorithm. By defining an objective function in terms of jitter ratio (JR) and underlying constraint condition derived from the spectrum of the sequence, the method is able to gradually reduce the searching-space guaranteed by a dynamic scaling factor in the mutation process of the DE algorithm. Four compatible sequences are given to make comparisons with the manual selected sequences as reported. The performance evaluated by the synthetic data justifies the feasibility of the present method.