Search

Nowadays, China has entered into an aging society; how to ensure safety in elderly care has drawn social attention. Through artificial intelligence and multi-information fusion research, combined with the applications of machine learning algorithms, internet of things devices and cloud computing, this paper presents a comprehensive, intelligent safety monitoring system for the elderly in the community and at home. The system collects the daily life data of the elderly through a series of sensors in an all-round, all-time, and non-intrusive manner, and realizes intelligent alarms for high-risk states such as falls, acute illness, abnormal personnel, and gas smoke for the elderly. Through the innovative research of human pose estimation and behavior recognition, and application of multi-sensor information fusion, the system can greatly reduce the occurrence or injury caused by safety incidents in senior care, bringing safe and healthy living environment for the elderly at homes and communities.

Retinal vascular function is complex, morphological structure varies from person to person, and is susceptible to vascular diseases and systemic vascular diseases. Its accurate segmentation is of great significance for disease diagnosis and identification. In this paper, a multi-scale matching filtering algorithm is proposed for the uneven size of retinal blood vessels. On the basis of the traditional singlescale Gaussian matching filter, multiscale Gaussian matched filters with two sizes are used to enhance grayscale images. Enhancement is performed, and the superimposed image is binarized using a twodimensional maximum entropy threshold segmentation algorithm. The algorithm is tested in the DRIVE database with sensitivity, specificity and accuracy of 0.803, 0.959, 0.981, respectively. Comparing with the traditional algorithm, the algorithm has high sensitivity, fast running speed and rich details of segmentation results.

Image outliers such as missing correspondences and large local deformations break the one-to-one pixelwise mapping between target image and moving image to be registered. Both traditional registration methods and deep-learning based deformable image registration methods fail to tackle this problem. This paper proposed an unsupervised globalto-local deformable registration network reinforced by joint saliency map to accurately, robustly and fast address the problem. The global-to-local network divided the overall learning of a complex mapping of image registration into a simpler global mapping learning and local residual mapping. The joint saliency map of the two images to be registered bidirectionally reinforced the whole network's forward estimation and back-propagation with uncertainty modeling and context-aware intelligence. The experimental results confirm the proposed method's performance advantages over the state-of-the-arts registration methods in the challenges image registration with missing correspondences and large local deformations.

In this paper, the classification and location of neuroblastoma in NMR images are realized by using Deep Neural Network(CNN) algorithm as the core technology. The module is integrated to realize the development of computer-aided diagnostic software. It is used to make up for the gap in the field of intelligent identification and accurate positioning of neuroblastoma in the current nuclear magnetic resonance detection technology, effectively reduce the work intensity of doctors reading films, and further promote the clinical application and technical development of nuclear magnetic resonance detection technology in the diagnosis of neuroblastoma.

Bowel sounds are one of the important physiological signals of the body,and different bowel sounds can reflect different gastrointestinal states.In this paper,long time bowel sound data is obtained with wearable full belly bowel sound recorder which is independent designed.After adaptive noise cancellation and wavelet threshold denoising,voice endpoint detection method based on short-time energy is used to identify effective bowel sounds.Experiments and results show that the sound recorder is simple and reliable.Through processing,analysis and endpoint detection algorithm,the recognition accuracy of effective bowel sounds is high,which has certain clinical practicality and research significance.

Numerical simulation is a powerful technology for photoacoustic imaging (PAI) in both theory studies and practical applications. In this paper, a simulation platform for PAI was designed and implemented based on Matlab. The simulation platform utilized finite element method (FEM) and k -space pseudospectral method to calculate the forward and inverse problem of PAI. And a graphical user interface (GUI) was realized. Structural design, work process and other operating details of the platform was also provided. By compared with theoretical temporal waveform of photoacoustic signal and reconstruction results of COMSOL, the validity and reliability was verified. And a reliable simulation tool was proposed for PAI.

Multi-angle plane-wave beamforming algorithm is the basis of ultra-fast ultrasonic imaging. It can be used to improve the imaging frame rate and resolution of traditional focused ultrasound. However, the existing multi-angle plane-wave technology can not satisfy the real-time imaging requirements due to the huge amount of computation required by CPU. In this paper, We proposed a parallel processing method to reduce the computation time based on compute unified device architecture(CUDA). Simulation analysis and contrast experiment were conducted to verify its performance. Experimental results show that the execution time based on GPU is much less than that based on CPU, thus the computational speed is accelerated significantly to satisfy the demand of ultrafast imaging.

Amblyopia is the common disease in the children's growth. There are some disadvantages in traditional treatment to amblyopia, such as poor compliance, easily leading to the healthy eye eyesight problem, the poor recovery of binocular and stereoscopy vision, etc. This paper put forward a new individual amblyopia treatment system to solve these problems. It combined pattern reversal visual evoked potential (P_VEP) feedback and dichoptic training theory and then developed the EEG data acquisition and synchronized trigger circuit using and I/O data collecting card, realized the raw EEG signal real-time sampling, extracted VEP waveform from the raw EEG data by using digital filtering algorithm, average superposition algorithm and wavelet an lgorithm, finally, the latency and peak of P_VEP wave shape were accurately measured. This system also present an individual treatment solution based on dichoptic training model by searching and obtaining the best spatial frequency of the treatment training picture and combining with 3D display technology.

Brain-computer interface (BCI) provides a new choice for people who lose communication ability, so the recognition of EEG has been paid attention. In this paper, wavelet packet transform (WPT) and transfer learning (TL) were used to classify right-hand and foot motion imagery tasks. Firstly, based on analyzing the channels and frequency bands closely related to event-related desynchronization (ERD), the EEG signals are decomposed by WPT. Then the relevant nodes were selected to calculate wavelet packet energy. Finally, TL was used to classify the BCI competition Ⅲ data IVa. The ideal classification result was obtained. The results show that the method is simple and effective, and it is valuable for online application of BCI.

In order to facilitate the doctor to obtain information more comprehensively and operate accurately from a three-dimensional perspective in interventional therapy, the paper uses a multi-view stereo reconstruction algorithm for three-dimensional reconstruction of medical implants in vitro. The effects of shooting and environmental conditions on reconstruction time and effect were analyzed. We use the convenient shooting device to collect multi-view images of the stent and open source software to acquire the point cloud data. By using reverse engineering software, the point cloud is processed interactively and the surface is generated. Eventually, the 3D model of the implant is reconstructed, which can be used for industrial production.

In Fluorescence Molecular Tomography (FMT), the reconstruction and fusion display of inner light source can be realized with the aid of CT data. A method is proposed in this paper to select permissible source region based on FMT image and CT data. The position and depth of light source can be estimated in FMT image. According to the coordinate relation of the two mode image, the light source position can be marked in CT data and mesh was produced on the marked CT data. Experimental results show that, the method can reduce the ill-posedness of the inverse problem and improve the accuracy of reconstruction of inner light source.

This paper presents a new, simple and fast algorithm of automated P wave detection in multi-lead ECG Signals. Range of QRS-T complex is detected firstly. Then QRS-T complex is eliminated. Final y, P wave’s onset and offset are detected by using the average of low-pass difference method and tangent method. Tangent method is always used to locate the offset of T Wave but it wil firstly be used to locate the P wave onset and offset in this paper. The proposed algorithm is tested by the annotated CSE database. Result shows that algorithm test result has a good consistency with BIS CSE annotation. Compared with the mean and standard deviation of P wave onset and offset, our algorithm and CSE annotation is-2.01 ms, 3.94 ms and 4.96 ms, 5.86 ms.

Ultrasonic plane wave imaging is one of the hot topics in the study of ultrasound imaging, but the plane wave ultra-high frequency ultrasound has brought a large amount of data. If these data are transmitted directly through the hardware, it will not only bring a great burden on the hardware, but also put a requirement for the ability of the hardware processing data. In order to reduce the problem caused by the high frame frequency and large amount of data, the data compression methods are used to reduce the amount of data acquisition and transmission. In this paper both lossy compression and lossless compression are studied, where LZW compression algorithm is chosen to test lossless compression effect, and based on the plane wave characteristics, the frequency domain cutting compression algorithm is developed as a lossy compression method. In this paper, the LZW algorithm is implemented in FGPA, and it can achieve a certain compression ratio after testing the ultra sonic simulation data. Under the condition of known fractional bandwidth, it is found that the main bandwidth can be preserved in the spectrum, while the rest of the frequency signal has little effect on the image. By testing the ultrasonic simulation data, the frequency domain cutting algorithm can achieve high compression ratio, and the difference between the original image and the reconstructed image is relatively smal . So from the point of the compression ratio, the frequency domain cutting algorithm is the primary choice of the ultrasonic data compression algorithm.

Optical Coherence Tomography(OCT) system may cause dislocations and distortions of the collected image volumetric?data when proceeding dynamic 3D imaging because of the involuntary movements of eyebal, which wil result in misdiagnoses during clinical diagnosis. The optic disk region of human eye has much more complex structure and further more blood vessels compared with macular region, which means the conventional image correction method aiming at macular region can not apply to optic disk region. In this paper, we propose a X direction artifact correcting method based on pixel row correlation matching algorithm of C-Scan, specifi c to the structural features of human eye optic disk region OCT images. Moreover, using this method, we can estimate whether the image sequences with motion artifacts have part image repetition or missing or not, and we can further confi rm the positions and numbers of repetition and missing with the help of Y direction 2D scan images. Model verifi cation indicates that this method can effectively rebuild the real image sequences and truly refl ect the physiology structure of retina.

At present, there is no accurate and quantitative methods for the determination of cardiac mechanical synchronism, and quantitative determination of the synchronization function of the four cardiac cavities with medical images has a great clinical value. This paper uses the whole heart ultrasound image sequence, and segments the left & right atriums and left & right ventricles of each frame. After the segmentation, the number of pixels in each cavity and in each frame is recorded, and the areas of the four cavities of the image sequence are therefore obtained. The area change curves of the four cavities are further extracted, and the synchronous information of the four cavities is obtained. Because of the low SNR of Ultrasound images, the boundary lines of cardiac cavities are vague, so the extraction of cardiac contours is still a challenging problem. Therefore, the ASM model information is added to the traditional level set method to force the curve evolution process. According to the experimental results, the improved method improves the accuracy of the segmentation. Furthermore, based on the ventricular segmentation, the right and left ventricular systolic functions are evaluated, mainly according to the area changes. The synchronization of the four cavities of the heart is estimated based on the area changes and the volume changes.

Objective: An optimal automatic selection method of permissible source region is proposed to reduce the ill-conditioned and ill-posed problems in the reconstruction of the light source in bioluminescence tomography., Methods: The 2D images captured by CCD are mapped into surface light irradiance distribution based on the light propagating model. The relation matrix between the source and light distribution is obtained by finite element method. Permissive source region is determined by using the automatic selection method proposed in this paper, and then Tikhonov regularization is applied to reconstruct the light source., Results: The center point distance between the optimal permissible source region and true source is 1.26 mm, and the center point error of the reconstructed light source and true source is 0.47 mm, the volume error is 9.13 mm3., Conclusion: The optimal permissive source region selection strategy is effective to locate the permissive source region close to the true source, and reduces the reconstructed error due to subjective orientation of permissible source region. This proposed method is the basis of high precision source reconstruction in bioluminescence tomography.

The traditional denoising algorithm for ultrasound images would lost a lot of details and weak edge information when suppressing speckle noise. A new denoising algorithm of adaptive threshold based on curvelet transform is proposed in this paper. The algorithm utilizes differences of coefficients' local variance between texture and smooth region in each layer of ultrasound image to define fuzzy regions and membership functions. In the end, using the adaptive threshold that determine by the membership function to denoise the ultrasound image. The experimental text shows that the algorithm can reduce the speckle noise effectively and retain the detail information of original image at the same time, thus it can greatly enhance the performance of B ultrasound instrument.

Motion segment and extraction from continuous signals is the premise of surface electromyography (sEMG) analysis. For the problem that sEMG energy threshold required repeated manual testing to set, this the paper established a this mathematical model of continuous actions based on Gaussian sEMG energy curve, in which the energy threshold was set according to the distribution of Gaussian signal section, and differentiated the action signals from no-action signals combined with energy comparison method. The experiment results showed the method can achieve continuous repetitive action segmentation, and compared with manual segmentation of the same signal, has a higher degree of coincidence.

Objective: Using the algorithm on the layered structure of the retina and quantitative analysis of the automatic segmentation technique is the key to the early diagnosis of glaucoma and other retinopathy on optical coherence tomography. Existing methods require high quality image and have low reliability. This paper used the improved complex nonlinear diffuse filtering and other methods to solve this problem., Methods: This paper includes algorithm such as automatic threshold, improved complex nonlinear diffusion filtering, morphological operations and peak detection. Use the method for the segmentation of 20 retinal layers images which acquired on the self-builder OCT system, the boundary of inner limiting membrane(ILM), outer nuclear layer(ONL), the photoreceptor segments(IS/ OS) and the RPE_ChCap layer are detected accurately. At last, the photoreceptor layer thickness is measured., Results: The results of segmentation and measurement are good corresponded with expert manual segmentation and measurements, retinal foveal measurements data is consistent with Zeiss Stratus OCT central retinal thickness 212+/-20 micro m., Conclusion: The algorithm proposed is prospective applied to clinical diagnosis of retinal diseases.

Evaluating the balanced capacity effectively is convenient to take some appropriate therapy measures to avoid some falling accidents. In the paper, we collect pressure data from the sole of feet at eight postures of standing and get the track of the man's center of gravity. Then we process these data with fast Fourier transform(FFT), after that, we can see the relation between amplitude and frequency of movement, and know very well shaking degree of standing, then get a technique data of evaluating man's balanced capacity objectively.

In this paper, experiments were designed for inducing neutral, terrified, excited, annoying emotions, and also low, middle, high, three levels of tension emotions of stress state, respectively. Based on the multi physiological signals generated by the subjects in emotions, such as heart rate and respiration rate and so on, we extracted features from these data which had been eliminated the baseline. Then the Particle Swarm Optimization method was adopted to optimize the features selection from the features of multi physiological signals, and combined with k-Nearest Neighbor algorithm, different emotions and varying degree tensions were classified. The result shows that the classification accuracy of the kNN method with SPO and baseline eliminated is better than the traditional kNN method.

In order to extract the outlines of macular edema from OCT images of macular, and estimate the volume of edema, we have to accurately segment the macular edema region. In this paper, an improved PCNN algorithm was proposed to conduct the above process. Combined with the adaptive base threshold, and the simplified neural network parameters, a binary image of macular edema was produced. According to the principle of maximum image entropy, the optimal number of iterations was determined as 8, which was evaluated by its misclassification rate. Simulation showed that the proposed algorithm could extract the macular edema region rapidly and accurately, providing the basis for further OCT image analysis.

This paper reviewed the dose calculation method for treatment planning system of multi-source stereotactic radiotherapy system with Gamma beam, and presented a fast method based on cone coordinate system and multi-core parallelization. The experiments show that the new method not only ensures the accuracy of dose calculation, but also greatly improves the speed, which can reach as 20 times as that of traditional method.

Three algorithms of electrical impedance tomography (EIT) are studied in this paper. The image resolution, anti-noise property and computation rapidity of the reconstruction algorithms are compared. As a result, it shows that back-projection algorithm has good anti-noise property, that NOSER algorithm generates images with good resolution, and that sensitivity matrix algorithm has moderate property.

When MR image's area is too small compared with the whole picture, the use of the current auto-window algorithms usually gets poor clarity and contrast. In order to address this problem, an improved auto-window algorithm is proposed in this paper and can solve the problem effectively and get clear and rich layers of MR images quickly and easily.

This paper introduces an adaptive filtering algorithm based on the LMS principle to inhibit the ventricular interference in the atrial epicardial mapping experiments.

Electromagnetic tracking is a method to track the position and space attitude of objects utilizing electromagnetic field. In this paper, the main methods during the development of Electromagnetic tracking are summarized and classified according to the magnetic field generation method, the number of magnetic sources and sensors used, the configuration and geometry of them. Tracking algorithm and calibration methods are also discussed. The critical issues during the development of electromagnetic tracking are analyzed, with the hope of helping the further research and application of electromagnetic tracking.

A method is presented in this paper which processes the human chest CT image with binary morphology theory, then marks and searches each connected domain to find chest contour and boundaries of lung and esophagus. Comparing with original CT image, the extracted boundaries have sufficient accuracy and human eyes cannot distinguish the error.

Based on the heart rate measurement method using time-lapse image of human cheek, this paper proposes a novel measurement algorithm based on Artificial Intelligence. The algorithm combining with fuzzy logic theory acquires the heart beat point by using the defined fuzzy membership function of each sampled point. As a result, it calculates the heart rate by counting the heart beat points in a certain time period. Experiment shows said algorithm satisfies in operability, accuracy and robustness, which leads to constant practical value.

This paper introduces the basic theory and arithmetic of Independent Component Analysis which is a novel technology developed for Blind Source Separation. The Fast ICA arithmetic is applied as an example in analysis and separation of artifacts in clinical multiple channel EEG. The experiment results testify that by using ICA method artifacts in EEG such as EOG and power line interference can be separated and removed effectively.

Multiscale pyramid image enhancement algorithm is an usual way to enhance the Digital Radiography (DR) images. However, the process of enhancement takes much of time because of the fine resolution of DR images. A new method of accelerating DR image enhancement based on Compute Unified Device Architecture (CUDA) is presented in this paper. This method completes a large amount of convolution operations in spatial domain involved in the multiscale pyramid image enhancement algorithm by using the Graphic Processing Unit (GPU). The experimental results show that the proposed method is very efficient for accelerating DR image enhancement.

EMG, This paper introduces a slope variability algorithm which uses the CU Database and the AHA Database. The sensitivity, specificity, positive predictivity and accuracy are calculated and these values are compared with the results from two earlier detection algorithms. The results show this algorithm is more effective and fast than other algorithms.

Ultrasound image has a lot of speckle noise, which brings great difficulties to the feature extraction, recognition and analysis. Especially in the edge extraction, the conventional extraction algorithms are difficult to achieve the desired results because of the speckle noise. To solve this problem, an algorithm based on the anisotropic diffusion equation is presented. The new algorithm combines the robust estimation and considers the characteristic of the speckle noise, so it can suppress the speckle noise effectively and be more robust, thus the edge details of the ultrasound image can be preserved even enhanced, which can provide effective safeguard for the following edge extraction. Furthermore, the paper proposes a new method to compute speckle scale coefficients automatically, which reduces the influence of the human beings, and enhances the stability of the algorithm.

In this paper, a T wave alternans detection system based on wireless and USB communication is presented. The ECG signal is amplified in the first step. The amplifier has a gain of 2000 and the CMRR is 80dB. Then the signal is sampled with a 12-bit AD converter. At last, the sample result is transmitted to the PC via wireless communication and USB bus. In the PC, the T wave alternans are detected using the spectral method. Results show that this system has high data transmission rate and real time data processing capability, and can realize the detection of T wave alternans.

This article used support vector machine (SVM) algorithm to recognize the particles in urine sediment in this paper. After feature extraction, cross-validation method and the contour chart of the accuracy were implemented to select the kernel function and the parameters of SVM, and according to the characteristics of SVM classifier and sample data, Multi-SVMs with two-level-classifier was successfully designed and A classification matrix was eventually obtained. The evaluation by using clinical data and comparative results with the artificial neural network have demonstrated that the proposed algorithm gets better results.

After extracting definition measurements by different focus functions from the images, a novel autofocus algorithm which combines them with principal component analysis (PCA) and considers the first principal component as the final measurement, is presented in this paper. The experiment results with 70 groups of images show that this algorithm increases the difference between the definition measurements of the images and provides higher focusing accuracy in comparison with other single ones.

In this paper, an algorithm for atrial premature beat classification with abnormal ECG rhythm analysis is presented. The classification algorithm identifies QRS morphology by template matching and recognizes abnormal ECG rhythm by the features of heart rate and ECG. Algorithm has been evaluated by the MIT-BIH arrhythmia database and the results show that the specificity and the sensitivity for atrial premature beat are 99.5% and 96.9% respectively. atrial premature, arrhythmia, beat classification

Combining the speed of two-dimensional detection with the precision of three-dimensional detection, an automatic algorithm based on high resolution CT images is proposed to identify nodules in this paper. Nodule candidates are extracted by a two-dimensional convergence index (CI) filter, then a three-dimensional Hessian matrix detection filter is introduced to reduce false positive lung nodules, Experiments show that the algorithm is effective with a sensitivity of 90% and the false positive lung nodules per slice is 0.33.

Based on chest images' features, this paper presents a retrieval method with domination color of partition, gray level co-occurrence matrix and gradient image shape invariant moments, as well as the fusion of them and the technique of relevance feedback. In Windows Vista environment with Visual C + +, a chest image retrieval prototype is implemented and the experiments have proved that this method has a better retrieval effect.

A new automatic selection approach of microorganism specific fragment combination is presented in this paper. Genetic algorithm is used to search optimal solution on the basis of classification ability of SNP combination, which is evaluated by the rough set theory. Other related experimental parameters are also been incorporated. Experimental results show that the method can find the best SNP combination pattern efficiently and accurately, which implies that it is a reliable approach to the genechip probe design.

A fast microscopic image mosaicing method is proposed in this paper by making a study of the mosaic methods and the characteristics of microscopic images. In the paper, invariant local features based on normalized moment of inertia (NMI) are used to select the matching points and calculate the spatial translation. The experimental results demonstrate that this algorithm can achieve fast, effective microscopic image mosaicing.

This paper proposes a tree-structured wavelet algorithm for brain CT texture feature extraction. This algorithm is used to analyse subimages automatically and to choose the optimum subimage for feature extraction. Experimental results show that this algorithm can improve the performance of brain CT texture feature extraction.

On account of different thickness of tissues absorbing different doses of X ray, it is difficult to display anatomical details of different tissues on a single screen of DR equipment. In order to resolve this problem, the self-adaptive reverse S mode transformation algorithm is presented in the paper, which can modify darker regions and brighter regions respectively, and improve the visibility of weakly details.

The paper presents the development and design of an ultrasonic Doppler fetal heart rate monitoring system based on DSP, and explains the realization of a algorithm for detecting fetal heart rate. Clinical practice has proved that this system is able to pick up the real-time fetal heart rate correctly and rapidly.

In this paper, a method based on the convex hull algorithm is presented for extracting modelling data from the locations of catheter electrodes within a cardiac chamber, so as to create a 3-D model of the heart chamber during diastole and to obtain a good result in the 3-D reconstruction of the chamber based on VTK.

The application of mathematical morphology to ECG signal processing is described in this paper and the prospects for the application of mathematical morphology are given too.

A CT 3-D image is reconstructed based on a lot of 2-D slice images. A new 3-D image reconstruction method, presented in this paper, is to use the helical scan continuity, sufficient condition of image reconstruction and the raw data from a few helical scan cycles to reconstruct,by a direct interpolation algorithm, 2-D images and then 3-D images.

This paper presents a machine learning method to select best geometric features for deformable brain registration for each brain location. By incorporating those learned best attribute vector into the framework of HAMMER registration algorithm, The accuracy has increased by about 10% in estimating the simulated deformation fields. At the same time, on real MR brain images, we have found a great deal of improvement of registration in cortical regions.

This paper presents a new deformable model using both population-based and patient-specific shape statistics to segment lung fields from serial chest radiographs. First, a modified scale-invariant feature transform (SIFT) local descriptor is used to characterize the image features in the vicinity of each pixel, so that the deformable model deforms in a way that seeks for the region with similar SIFT local descriptors; second, the deformable model is constrained by both population-based and patient-specific shape statistics. At first, population-based shape statistics plays an leading role when the number of serial images is small, and gradually, patient-specific shape statistics plays a more and more important role after a sufficient number of segmentation results on the same patient have been obtained. The proposed deformable model can adapt to the shape variability of different patients, and obtain more robust and accurate segmentation results.

In this paper, a fuzzy neural network (FNN) is proposed for fusing the anesthesia information, and realizing the monitoring of the depth of anesthesia (DOA). EEG data from 31 patients undergoing general anesthesia with different anesthetic agents, and Kc complexity (Kc), approximate entropy (ApEn) were extracted and the fuzzy neural network was trained by 25 samples, and tested by the other 6 samples. The results show that the outputs of the fuzzy neural network whose inputs were Kc and ApEn obtained under the awake state and asleep state, exist obvious difference. It can be regarded as an quantitative index to estimate DOA.