Search

With the booming development of medical information technology and computer science, the medical services industry is gradually transiting from information technology to intelligence. The medical knowledge graph plays an important role in intelligent medical applications such as knowledge questions and answers and intelligent diagnosis, and is a key technology for promoting wise medical care and the basis for intelligent management of medical information. In order to fully exploit the great potential of knowledge graphs in the medical field, this paper focuses on five aspects: inter-drug relationship discovery, assisted diagnosis, personalized recommendation, decision support and intelligent prediction. The latest research progress on medical knowledge graphs is introduced, and relevant suggestions are made in light of the current challenges and problems faced by medical knowledge graphs to provide reference for promoting the wide application of medical knowledge graphs.

Brain-computer interface (BCI) provides a direct communicating and controlling approach between the brain and surrounding environment, which attracts a wide range of interest in the fields of brain science and artificial intelligence. It is a core to decode the electroencephalogram (EEG) feature in the BCI system. The decoding efficiency highly depends on the feature extraction and feature classification algorithms. In this paper, we first introduce the commonly-used EEG features in the BCI system. Then we introduce the basic classical algorithms and their advanced versions used in the BCI system. Finally, we present some new BCI algorithms proposed in recent years. We hope this paper can spark fresh thinking for the research and development of high-performance BCI system.

This paper is aimed to fulfill a prototype system used to classify and retrieve retinal image automatically. With the content-based image retrieval (CBIR) technology, a method to represent the retinal characteristics mixing the fundus image color (gray) histogram with bright, dark region features and other local comprehensive information was proposed. The method uses kernel principal component analysis (KPCA) to further extract nonlinear features and dimensionality reduced. It also puts forward a measurement method using support vector machine (SVM) on KPCA weighted distance in similarity measure aspect. Testing 300 samples with this prototype system randomly, we obtained the total image number of wrong retrieved 32, and the retrieval rate 89.33%. It showed that the identification rate of the system for retinal image was high.

This paper presents an automatic calculation method for angiography image, which enables programs to intellectively acquire several parameters of blood vessels, such as contours, segments, widths, etc. and then intellectively identify the angiostenosis parts. This method is a kind of automatic optic inspection (AOI) technology. Blood vessels usually distribute as curves and have a fastening direction. According to this feature, the approach performs inspection automatically using improved Steger algorithm, which firstly computes the convolution between image and Gaussian function kernel, and then computes second order Taylor expansion at eac pixel. And further the eigenvalues and eigenvectors of Hessian matrix are calculated on each pixel to obtain the direction of lines and local maximum of second derivative at that point. Hysteresis threshold and directional connection operators are then used to generate blood vessel skeleton. Finally we can compute the blood vessel widths for every sub-pixel object points on blood vessel curve. For given digital X-ray images of hearts with blood vessel local straitness, experiments showed that this method had the ability of getting all the data we need and could find the local confined parts in blood vessels. This approach is proved to have a good effort for angiography images, and it has some advantages such as fast speed, high accuracy, good robustness and no need for human interventions. It could also be a promising computer aided diagnosis method.

Colon cancer is one of the leading causes of cancer-related deaths, while the virtual colonoscopy plays an important role in the early detection of colon cancer. Effective colon segmentation is the basis of virtual colonoscopy. Oral contrast agent can reduce the use of laxatives, but it can to bring some difficulties in the colon segmentation. Because of the partial volume effect, the CT value of air-tagging boundary span a wide range, which are not easily eliminated. We present an effective method of colon segmentation in this paper. Firstly, we used adaptive threshold method to extract air area. Then we segmented tagging region and removed air-tagging boundary by segment region growing method based on statistical information and morphological operation. The experiments showed that the agreement between the results by manual and by orithm segmentation reached 93.4%.

At present, ray-casting algorithm is the most widely used algorithm in the field of medical image visualization, and it can achieve the best image quality. Due to large amounts of computation like sampling, gradient, lighting and blending calculation, the cost of ray-casting algorithm is very large. The characteristic of Streaming single instruction multiple datum extensions (SSE) instruction--supporting vector computation--can satisfy the property of ray-casting algorithm well. Therefore, in this paper, we improved the implementation efficient significantly by vectorization of gradient, lighting and blending calculation, and still achieved a high quality image at the same time.

The automatic detection and classification of EEG epileptic wave have great clinical significance. This paper proposes an empirical mode decomposition (EMD) and support vector machine (SVM) based classification method for non-stationary EEG. Firstly, EMD was used to decompose EEG into multiple empirical mode components. Secondly, effective features were extracted from the scales. Finally, the EEG was classified with SVM. The experiment indicated that this method could achieve good classification result with accuracy of 99 % for interictal and ictal EEGs.

Pattern selection plays an important role in data mining and pattern recognition, especially for large scale bioinformatic data. There are many problems in this field, such as algorithm complexity and numbers of the best feature subset. In this paper, we propose a new pattern selection algorithm, carrying out pattern selection base on Mutual Information (MI). Pattern subset evaluation index was studied to ensure the best feature subset. To pattern selection, algorithm bases on the correlation of patterns and label, as well as the redundancy of each pattern. Neurofuzzy Pattern Subset Evaluation Index was researched to make sure which is the best subset for our pattern subset evaluation. To verify the effectiveness of our method, several experiments are carried out on the data of gene expression of mouse from Leiden University and UCI datasets. The experimental results indicated that our algorithm achieved better results in the complexity and accuracy.

This paper is aimed to study a method and feasibility of early diagnostic system using hypoxic ischemic encephalopathy (HIE) in the newborns. Fifteen non-invasive indicators with high sensitivity and specificity were selected for the early diagnosis of HIE on the basis of related researches from the literature and the researches in our laboratory. The diagnostic test was done with 140 cases with the HIE, using the fussy BP neural network experiment system. The initial results showed that the accuracy rate was 100% for the training set and 95% for the testing set, and the error rate was 5%. The data suggested that the fuzzy back-propagation neural networks, with the clinical comprehensive indicators, exhibited a high accuracy for the early diagnosis of HIE. This method provides an objective and convenient new way for the early clinical diagnosis of the HIE.

The method for segmenting malignant halo of malignant breast tumor in ultrasound image is useful in providing evidence for the differential diagnosis of breast tumor. In this respect, we adopt an improved anisotropic diffusion filtering method to preprocess the breast tumor ultrasound image, and then apply an improved LBF model with combination of Otsu and morphology methods to extract internal and external contours for obtaining malignant halo based on LBF model proposed by Li. We compare our data of malignant halo with doctor's manual-sketched malignant halo, and make quantitative analysis. The result shows that the malignant halo segmented by the proposed methods in this paper is in accordance with the manual-sketched malignant halo.

In this paper, an image mosaic algorithm based on feature points was proposed by making a study of the image mosaic methods and the characteristics of pathological image. Points of interest were firstly extracted by Moravec operator in this method, and the corresponding feature matches of the original images were achieved by correlation coefficient. The parameters of affine transformation were calculated by the matching feature points, and weighted average method was used to fuse images. The experimental results show that, as for the pathological image with translation and rotation, the algorithm can extract the matching feature points accurately and realize the image seamless mosaic.

Ultrasound based tissue thermal lesion non-invasive detection is of great significance in high intensity focused ultrasound (HIFU) clinical application. In this paper, we propose a sub-pixel method to quantify the ultrasound image change caused by HIFU as correlation-distance. The support vector machine (SVM) was trained by using correlation distance as samples, and the recognition effect was tested. Results showed that sub-pixel cross-correlation vector field could reflect the ablation lesions position. SVM based classification method can recognize HIFU beam lesion degree effectively.

In allusion to non-rigid registration of medical images, the paper gives a practical feature points matching algorithm--the image registration algorithm based on the scale-invariant features transform (Scale Invariant Feature Transform, SIFT). The algorithm makes use of the image features of translation, rotation and affine transformation invariance in scale space to extract the image feature points. Bidirectional matching algorithm is chosen to establish the matching relations between the images, so the accuracy of image registrations is improved. On this basis, affine transform is chosen to complement the non-rigid registration, and normalized mutual information measure and PSO optimization algorithm are also chosen to optimize the registration process. The experimental results show that the method can achieve better registration results than the method based on mutual information.

In the field of computational molecule biology, there is still a challenging question of how to detect non-coding RNA gene in lots of unlabeled sequences. Generally, the methods of machine learning and classification are employed to answer this question. However, only a limited number of positive training samples and unlabeled samples are available. The negative samples are difficult to define appropriately, yet they are necessary for usual learning-then-classification method. The common way for most of the existing non-coding RNA gene finding methods is to produce a number of random sequences as negative samples, which may hold some characteristic of positive sample sequences. Consequently, the contrived uncertain factor was introduced and the performance of methods was not good enough. In this paper, Support Vector Data Description (SVDD) is in use for to learning and classification as well as for detecting non-coding RNA gene in lots of unlabeled sequences, and the k-means clustering algorithm is employed before SVDD training to deal with the high flase positive fault in the result of SVDD. The training samples (target samples) are non-coding RNA genes validated by experiment. Moreover, appropriate features were constructed by Principal Component Analysis (PCA). The effectiveness and performance of the method are demonstrated by testing the cases in NONCODE databases and E. coli genome.

Usually all heartbeats contribute to the calculation of electrocardiogram (ECG) waveform parameters, such as amplitude and interval, in which those low signal-to-noise-ratio (SNR) heartbeats actually produce a negative effect. In this paper is presented an approach for measuring ECG waveform parameters using template-based methods. Compared with Karhunen-Loeve (KL) template, average template has an advantage over KL template both on speediness and accuracy. Experimental results proved that this approach improved the statistical precision of ECG waveform parameters. In conclusion, this method makes the result of ECG auto analysis more accurate and reliable.

This paper addresses the predicting problem of peritoneal fluid absorption rate(PFAR). An innovative predicting model was developed, which employed the improved genetic algorithm embedded in neural network for predicting the important PFAR index in the peritoneal dialysis treatment process of renal failure. The significance of PFAR and the complexity of dialysis process were analyzed. The improved genetic algorithm was used for defining the initial weight and bias of neural network, and then the neural network was used for finding out the optimal predicting model of PFAR. This method utilizes the global search capability of genetic algorithm and the local search advantage of neural network completely. For the purpose of showing the validity of the model, the improved optimal predicting model is compared with the standard hybrid method of genetic algorithm and neural network. The simulation results show that the predicting accuracy of the improved optimal neural network is greatly improved and the learning process needs less time.

The electrocardiograph signal presents the sparseness. In this paper, a new method of extracting the Fetal Electrocardiograph (FECG) is proposed, which uses the Blind Source Separation(BSS) of Sparse Signal. Both the Wavelet Analysis method and a new scheme named Vanish Circle are also involved in order to avoid the influence on the signal separation, because the ECG signals are not completed sparse signals. Some experiments show that the new method is effective.

The chaos of heart rate variance (HRV) is identified by surrogate data method in this paper. Firstly, the main principle of chaotic identification is introduced, in which the median absolute error (MAE) of one-step prediction for HRV is set as the statistic. The algorithm is checked with a known chaotic system response and a colored noise signal. Then, some typical healthy and unhealthy HRVs are analyzed with the method of surrogate data, and some characteristic parameters from this method are compared.

The cough sound is a very important symptom of well over 100 kinds of diseases. Cough sound analysis can provide much information which is useful for diagnosing. Detecting the frequencies and intensity of cough can evaluate the efficiency of therapy quantitatively. In this paper, we put forward an algorithm for cough sound recognition. We first decompose signals with wavelet transform and calculate the normalized energy at each time-frequency point. Then we obtain the normalized energy distribution statistically. After that, we pick out the time-frequency points maximizing a certain discriminant measure of normalized energy distribution between cough sound and non-cough sound, and then we use the normalized energy belonging to these time-frequency points as the inputs of Linear discriminant analysis/Generalized singular value decomposition (LDA/GSVD) classifier. The experimental results show that the classification accuracies achieved by using the algorithm is about 85%, and the computation complexity is low.

It is very difficult to use the conventional methods that are based on area and on border for extracting a region of intered such as liver tumor region with vague and irregular boundaries. This paper introduces Poisson matting that uses transparency (alpha value) as self-property of image and seeks its best result value to achieve the aim of extracting object. Meanwhile, this method is applied to the liver tumor CT picture experimental results. It makes improvement in two steps: computing image source a value; calculating foreground image F and background image B. Consequently, successful result is obtained.

The shape of dental arch for orthodontic diagnosis and treatment is of great significance. This paper presents an automated method for detecting the dental arch form. Firstly, 3D teeth data model is retrieved by the 3D-optical measuring system. Secondly, the occlusal plane is computed by interactively picking up four feature points. Thirdly, the feature point set is filtered by the rule and two-step curve fitting method is used to obtain the dental arch form. Finally, some examples are tested in this work and the results demonstrate that the proposed algorithm is effective and feasible.

The interference of noise and the weak edge characteristic of symptom information on medical images prevent the traditional methods of segmentation from having good effects. In this paper is proposed a boundary detection method of focus which is based on dyadic wavelet transform and active contour model. In this method, the true edge points are detected by dyadic wavelet transform and linked by improved fast active contour model algorithm. The result of experiment on MRI of brain shows that the method can remove the influence of noise effective and detect the contour of brain tumor actually.

Image guided surgery (IGS) means that computers are used for modeling the position of patients and providing the surgeons with image guidance for invasive procedures. In this paper, we introduce the composition and the work flow of IGS, analyze the process and principle of Registration, which is regarded as a process to combine the actual position of patients and the visual position of models. Seeing that the precision of the registration determines the precision of the IGS system directly, we conduct a study on the registration algorithm precision, feasibility as well as robustness. Combined with the actual requirements in clinical IGS, we also analyze the influence of changes in the number and position of markers on registration precision.

Recognition by gait is a new field for the biometric recognition technology. Its aim is to recognize people and detect physiological, pathological and mental characters by their walk style. The use of gait as a biometric for human identification is promising. The technique of gait recognition, as an attractive research area of biomedical information detection, attracts more and more attention. In this paper is introduced a survey of the basic theory, existing gait recognition methods and potential prospects. The latest progress and key factors of research difficulties are analyzed, and future researches are envisaged.

Automatic speaker gender identification based on voice feature is an important task in voice processing and analysis fields. In this paper non-linear parameters such as fractal dimension are applied to be one part of feature space for improving the ability of describing speaker gender feature through conventional linear parameters method. Pitch is picked using lifting scheme, and audio fractal dimension is extracted. Then based on Takens theory, the time delay method is used to reconstruct the phase space of fractal dimension sequence. And fractal dimension complexity is obtained by calculating Approximate Entropy. Three dimension feature vectors, including the pitch, the fractal dimension and the fractal dimension complexity, are applied to speaker gender identification. Experiment results show that through adding non-linear parameters, compared with the linear parameter using one dimension only such as pitch, the proposed method is more accurate and robust, and thus provides a new way for speaker gender identification.

In this paper, a new calculating method of generalized dimension, based on boxes-coupling principle, is proposed to overcome the edge effects and to improve the capability of the speech endpoint detection which is based on the original calculating method of generalized dimension. This new method has been applied to speech endpoint detection. Firstly, the length of overlapping border was determined, and through calculating the generalized dimension by covering the speech signal with overlapped boxes, three-dimension feature vectors including the box dimension, the information dimension and the correlation dimension were obtained. Secondly, in the light of the relation between feature distance and similarity degree, feature extraction was conducted by use of common distance. Lastly, bi-threshold method was used to classify the speech signals. The results of experiment indicated that, by comparison with the original generalized dimension (OGD) and the spectral entropy (SE) algorithm, the proposed method is more robust and effective for detecting the speech signals which contain different kinds of noise in different signal noise ratio (SNR), especially in low SNR.

To solve the problem of imprecise positioning of feature point and of the feature data redundancy in facial expression recognition by active appearance models (AAM), the automatic adjustment of initial model for AAM fitting is proposed in this paper. The specific aims are to improve the precision of positioning and to more effectively reflect the variation of expressions by acquired features. The problem of feature selection is resolved by adopting quadratic mutual information and reducing the feature dimension. The support vector machine (SVM) classifier is used for expression recognition. The experimental results on CAS-PEAL facial expression database show that the proposed method effectively improves the performance of facial expression recognition, the maximum recognition rate being 83.33%.

How to recognize and diagnose heart diseases, as one of the most frequent issues, is a serious problem. ECG automation recognition is not only the target of medicine, but also the work of computer science. In this paper, the approaches to ECG recognition are reviewed, and their characters are analyzed. With the background of wearable wireless ECG monitor, on the basis of experts' experience in ECG recognition, the imagery thinking-based new method, in which morphology parameters are considered, is presented.

In this paper, BP neural network is used to segment whole body bone SPECT image so that the lesion area can be recognized automatically. For the uncertain characteristics of SPECT images, it is hard to achieve good segmentation result if only the BP neural network is employed. Therefore, the segmentation process is divided into three steps: first, the optimal gray threshold segmentation method is employed for preprocessing, then BP neural network is used to roughly identify the lesions, and finally template match method and symmetry-removing program are adopted to delete the wrongly recognized areas.

Non-invasive manometry of esophageal varices is a cynosure of researchers. This paper develops a method based on computer vision. Experiments results reveal that correct pressure value can be got quickly.

As is well known that the quality of training samples directly influence the recognizing ability of neural network. In this paper, we introduce a method for solving the problem of how to classify the pattern of forearm by obtaining typical samples. At first, the original samples were pretreated by using the membership class function that can improve the quality of cluster sample. Then, the center of clustering could be gained by using the method of clustering and the typical sample was obtained. Based on this method, we can get the typical sample that corresponds with the movements of stretch of arm and fold of arm. We can make them as the training sample of the BP network to classify the pattern of forearm. The experiment indicates that this measure can improve the point of identification.

In present, the most basically used parameters for speaker identification are linear predictive coding (LPC) parameter, Mel frequency cepstrum coefficient(MFCC), etc. First in this paper only MFCC was used as the parameter and then Lempel-Ziv Complexity was combined with MFCC as parameters. The text-dependent recognition rate of 50 speakers increased from 42% to 80% and the text-independent recognition rate of 50 speakers increased from 60% to 72%. This test shows that Lempel-Ziv complexity, as a new parameter, can be applied to speaker identification.

Some noises still exist in the single-trial averaged visual evoked potentials (VEP), so further extraction of the above results is of significance. Independent component analysis (ICA)can separate the sources from their mixtures and make the output statistically as independent as possible; it can remove noises effectively. In this paper, the principle, experiment analyses and results of ICA based on quasi-Newton iteration rule for VEP feature extraction are introduced, It is compared with the fixed-point FastICA algorithm. The experiment results show that the provided algorithm may reinforce signals effectively and extract distinct P300 from the single-trial averaged VEP. It is of good applicability.

One of the applications of cDNA microarrays is to recognize the class and subclass of diseases such as cancers on the basis of statistical pattern classification methods using gene expression data. In this paper, we apply 2000 genes expression dataset provided by Affymatrix Company: 40 samples of intestine cancer tissue and 22 samples of normal tissue. We compare the performance of four pattern classification methods based on different feature selection methods. These pattern classification methods include: Fisher linear discriminate, Logit nonlinear discriminate, the least distance and K-nearest neighbor classifier. The results show firstly that four pattern classifiers based on the feature selection methods of t-test and classification tree all have better performance than those based on the stochastic feature selection methods, secondly that K-nearest neighbor classifier has the best performance, thirdly that both the least distance classifier and K-nearest neighbor classifier have better generalization, fourthly that four classifiers are less sensitive to the composition of samples.

It is very important to recognize arrhythmia in clinical electrocardiography (ECG) analysis. The fundamental of beat pattern recognition is presented in this paper. Various prevalent methods for arrhythmia recognitiion are categorized and summarized, based on which the advantages and disadvantages among the methods are compared, and the main problems are discussed in depth. At last, the development trend of arrhythmia recognition technology is pointed out.

This paper reviews the methods and research advances of recognition and classification of heart sounds. The methods include the statistical techniques and artificial neural network. It also discusses the main problems in the methods.