Kang-Ping Lin

Automatic breast region extraction from digital mammograms for PACS and telemammography applications

High spatial resolution results in very large digital mammogram file sizes. For telemammography, and picture archiving and communication systems, the large file issue introduces technical difficulties in image transmission, storage, and display. We propose extracting the breast region from the mammogram to reduce the image file size. The challenge is on how to faithfully extract breast regions from digital mammograms generated from different types of acquisition systems that contain various imaged compositions. We report an algorithm to automatically identify the orientation of breast region and extract the breast region from mammograms. Breast regions extracted from full-field digital mammograms reduce file sizes by three to five folds.

Classification of QRS pattern by an associative memory model

A feature-extraction method based on linear prediction for classification of QRS in an associative memory model is described. The feature extraction process converts each QRS pattern to a pulse-code train that describes only -1, 0, and +1 states. In order to recognize the feature of a QRS pattern, a two-layer feedforward neural net model is provided. The model shows the operation of each input node as well as a real neurons three typical states: resting (0), excitatory (+), and inhibitory (-1). The model performs well for arrhythmia detection.<<ETX>>

ECG analysis based on Hilbert transform descriptor

The authors used the Fourier descriptors (FD) of an electrocardiogram (ECG) and a synthetic vectorcardiogram (SVCG) to classify normal beat and premature ventricular contraction (PVC). They find that the magnitude of these descriptors decreases rapidly for higher-order harmonics, and the major shape information of SVCG is contained in a few of these descriptors. The distinction between a normal beat and PVC is very clear. They can efficiently classify normal beats and PVCs by just a few Fourier coefficients. The original SVCG can be reconstructed from few descriptors, results in a significant data compression. Five files of the MIT-BIH arrhythmia database were used for evaluation of this method. The accuracy of PVC detection is 98.77%.<<ETX>>

Statistical-based sub-band filtering technique for digital mammogram enhancement

Early detection and diagnosis of breast cancer markedly increases survival rate. Digital mammography is believed to help breast image experts to detect breast cancer early. Accurate diagnosis also depends on the quality of the image presented to the experts. This study addresses the quality improvement of the image. A statistically based sub-band filtering method is applied to enhance the mass and calcification shown in the digital mammograms by inhibiting noise. The method is based on sub-band transformation due to its decomposition characteristic and includes two steps: noise inhibition and boundary enhancement. Contrast ratios and frequency responses are measured to evaluate the enhancement performance and the distortion affect, respectively, to validate the effectiveness of the proposed technique. By comparing the enhancement performance of proposed and conventional methods, a phantom mammogram image that consists of similar mammographic microcalcifications, breast gland and well-circumscribed mass is designed for simulation experiment. Moreover, the real mammograms with microcalcifications are also applied on representing the efficient enhancement ability by the proposed method in this paper. Results in this study demonstrate that the proposed method improves the quality of the image more than other enhancement methods, according to these two criteria. The comparison results show that not only the image quality is improved but also within less image distortion.

Pattern recognition technique to QRS complex classification

A pattern recognition method is presented for QRS-complex classification. The linear prediction method is used to extract feature vectors of ECG signal, and the error correction rule is used to obtain the exact weight vector by training some prototype feature vectors. Then, every feature vector of the QRS-complex is decided in each class by a linear discriminant function. The MIT-BIH database was used to evaluate the present algorithm, and the results of classification have precisions of over 99%.<<ETX>>

A real-time feature extraction method for PVC detection in abedside monitor

Abstract This paper describes a fast and very efficient feature extraction method for discrimination of QRS and Premature Ventricular Contraction (PVC) beats in a microprocessor‐based bedside monitoring system. It converts each QRS and PVC beat into a positive‐pulse waveform by signal preprocessing. Two characteristic factors, the positive‐pulse and the pulse duration, are calculated when the onset and end points of each pulse have been detected by threshold detection. The prominent feature is extracted from a product of these two factors. This algorithm has been examined using 40 different patients’ electrocardiograph (ECG). The accuracy of QRS detection was 99.3 percent in the tests performed. The identification sensitivity of PVC was 81.2 percent with 18‐ different arrhythmia patients.

Adaptive noise reduction for pulmonary artery blood pressure

An application of time-sequenced adaptive filtering structure to reduction of the pulmonary artery pressure artifact is presented. The authors use the pulmonary artery pressure (PAP) signal for the input and an rectified sine wave signal contained correlated information (but not identical in waveshape) for the reference input. A filtered PAP signal from beat to beat provides stable systolic and diastolic pressure information. This filtering system is applicable to a microprocessor-based bedside monitor which requires simultaneous PAP monitoring.<<ETX>>

Feature analysis and extraction of tumor vascular structure on microscopic images

Tumor growth induces angiogenesis that results in immature/dysfunctional vessels. The purpose of this study is to demonstrate feature extraction and analysis of tumor vascular structure based on non-fluorescent microscopic images for long-term investigation. The averages of angle, and radius of microvasculature on the surface of the tumor were measured and analyzed at different weeks for each tumor. In the t-test results, we report statistical analyses which confirm at different significance levels, radius have significant effects in growing a tumors vascular network. No difference in average angle between parent vessel and children vessel by t-test was found.

Finite-Difference Time-Domain Simulation of Localized Surface Plasmon Resonance Adsorption by Gold Nanoparticles

Using optical sensors to transform light-matter interaction into optical signal has become more and more popular. This is especially true for the fields that require ultrafast responsibility and remote sensing, such as environmental monitoring, food analysis and medical diagnosis. Among numerous optical sensors, plasmonic nanosensors are of great promise due to their spectral tunability and good adaptability to modern nanobiotechnologies. Localized surface plasmon resonance (LSPR) is the electromagnetic resonance of conducting electrons on metal surface, and it is very sensitive to the variation of environmental refractive index. The LSPR is considered as a useful sensing parameter that provides very good biochemical information. The SPR absorption peak also can be adjusted by changing the nano structure on the LSPR biological sensor chip. In this study, Finite-Difference Time- Domain (FDTD) was applied to simulate the LSPR absorption peak. Four model parameters were modified to study the LSPR sensing sensitivity: (a) the incident light wavelength, (b) the diameter of nanoparticle, (c) the spacing among nanoparticles, and (d) the height of nanoparticle. The simulation results show that 860nm is the best wavelength for the LSPR adsorption measurement. The optimal diameter of nanoparticle is 150nm, and the nanoparticle spacing is 90nm. Higher nanoparticle height provides higher sensitivity, but it also depends on the process capability. The FDTD simulation can be a useful tool to design a LSPR nanoparticle biosensor.

An improved method for velocity estimation of red blood cell in microcirculation

This paper presents a coarse-to-fine combined method for dealing with large displacement situations caused by low speed of frame rate in microscopic video sequences. Motion image estimation method utilizes the modified block matching method based on image warping to perform a wide range of changes in the amount of search comparison, and then using the optical flow method to fine adjustment pixel by pixel, to complete the overall precision of the estimation. In the evaluation experiment, we have compared both current optical flow and proposed method by testing them with simulated vessel images, the results of the estimation is quite accurate.