Emotion can reflect the psychological and physiological health of human beings, and the main expression of human emotion is voice and facial expression. How to extract and effectively integrate the two modes of emotion information is one of the main challenges faced by emotion recognition. In this paper, a multi-branch bidirectional multi-scale time perception model is proposed, which can detect the forward and reverse speech Mel-frequency spectrum coefficients in the time dimension. At the same time, the model uses causal convolution to obtain temporal correlation information between different scale features, and assigns attention maps to them according to the information, so as to obtain multi-scale fusion of speech emotion features. Secondly, this paper proposes a two-modal feature dynamic fusion algorithm, which combines the advantages of AlexNet and uses overlapping maximum pooling layers to obtain richer fusion features from different modal feature mosaic matrices. Experimental results show that the accuracy of the multi-branch bidirectional multi-scale time sensing dual-modal emotion recognition model proposed in this paper reaches 97.67% and 90.14% respectively on the two public audio and video emotion data sets, which is superior to other common methods, indicating that the proposed emotion recognition model can effectively capture emotion feature information and improve the accuracy of emotion recognition.
Vascular perfusion distribution in fibroids contrast-enhanced ultrasound images provides useful pathological and physiological information, because the extraction of the vascular perfusion area can be helpful to quantitative evaluation of uterine fibroids blood supply. The pixel gray scale in vascular perfusion area of fibroids contrast-enhanced ultrasound image sequences is different from that in other regions, and, based on this, we proposed a method of extracting vascular perfusion area of fibroids. Firstly, we denoised the image sequence, and then we used Brox optical flow method to estimate motion of two adjacent frames, based on the results of the displacement field for motion correction. Finally, we extracted vascular perfusion region from the surrounding background based on the differences in gray scale for the magnitude of the rich blood supply area and lack of blood supply area in ultrasound images sequence. The experimental results showed that the algorithm could accurately extract the vascular perfusion area, reach the precision of identification of clinical perfusion area, and only small amount of calculation was needed and the process was fairly simple.
Medical image fusion realizes advantage integration of functional images and anatomical images. This article discusses the research progress of multi-model medical image fusion at feature level. We firstly describe the principle of medical image fusion at feature level. Then we analyze and summarize fuzzy sets, rough sets, D-S evidence theory, artificial neural network, principal component analysis and other fusion methods' applications in medical image fusion and get summery. Lastly, we in this article indicate present problems and the research direction of multi-model medical images in the future.
ObjectiveTo study the preservation effect of true bone ceramics (TBC) prepared by high-temperature calcination of bovine bone on alveolar ridge of canine extraction socket.MethodsSix healthy Beagle dogs (aged 1.5-2 years) were selected to extract the second and fourth premolars of both mandibles and the second premolars of the maxilla. The left extraction socket was implanted with TBC as the experimental group, and the right side was implanted with the calcined bovine bone (CBB) as the control group, to observe the alveolar ridge preservation effect. Three dogs were euthanized after general observation at 1 and 6 months after operation respectively. After separating the maxilla and mandible, cone beam CT (CBCT) was performed to measure the average gray value of the graft site and the adjacent reference area (the area between the roots of the adjacent third premolar) and calculate the gray scale ratio between the bone graft site and the reference area. Histological observation was made on the bone graft site to evaluate the new bone formation.ResultsGeneral observation showed that the wounds of both groups were basically healed at 2 weeks after operation, and the bone graft materials were not exposed. The wounds healed well at 1 and 6 months after operation without swelling. The results of CBCT showed that the residual material was found in both groups at 1 month after operation, and no significant residual material was found in both groups at 6 months after operation, and the alveolar ridge height of the bone graft area was not significantly reduced. There was no significant difference in the bone mineral density between the experimental group and the control group. The gray scale ratios of the experimental group at 1 month and 6 months after operation were 0.97±0.14 and 0.93±0.06, respectively, and were 0.99±0.16 and 0.94±0.05 in control group, showing no significant difference between the two groups (t=?1.030, P=0.333; t=?0.770, P=0.466). HE staining observation showed that a large number of bone graft materials did not degrade and new bone formed around the grafts in both groups at 1 month after operation; the bone graft materials were absorbed and a large number of new bones were formed in both groups at 6 months after operation.ConclusionTBC can maintain bone mineral density and have good osteoconductivity in the alveolar ridge site preservation experiment of dogs, and can be used for alveolar ridge site preservation.
ObjectiveTo observe the changes of macula in patients with high myopia after phacoemulsification. MethodsIn 20 patients with high myopia with ocular axial length≥27 mm, optical coherence tomography (OCT) was performed on the operative and contralateral eyes 1 week before and after monocular phacoemulsification, respectively, and the OCT images of macula of the operative eyes were observed and compared.ResultsOne week before and after phacoemulsification, the mean macular fovea thickness of the patients with high myopia was (131.6±16.37) μm and (189.75±45.69) μm, respectively, with a significant difference (t=2.805, P=0.01). Simultaneously, the mean macular fovea thickness of the contralateral eyes was (133.5±15.12) μm and (133.5±14.63) μm, respectively, with a non-significant difference (t=1.367, P=0.853). In 20 operative eyes 1 week after phacoemulsification, 3 had vitreous strand around the macula with retinal thickening, 1 had retinoschisis in macular area, and 2 had obvious retinal thickening with slight retinal edema.ConclusionRetinal thickening occurs in the patients with high myopia after phacoemulsification. Traction of retina by vitreous strand or subclinical retinoschisis may occur in some patients.(Chin J Ocul Fundus Dis, 2005,21:90-92)
Objective To automatically segment diabetic retinal exudation features from deep learning color fundus images. Methods An applied study. The method of this study is based on the U-shaped network model of the Indian Diabetic Retinopathy Image Dataset (IDRID) dataset, introduces deep residual convolution into the encoding and decoding stages, which can effectively extract seepage depth features, solve overfitting and feature interference problems, and improve the model's feature expression ability and lightweight performance. In addition, by introducing an improved context extraction module, the model can capture a wider range of feature information, enhance the perception ability of retinal lesions, and perform excellently in capturing small details and blurred edges. Finally, the introduction of convolutional triple attention mechanism allows the model to automatically learn feature weights, focus on important features, and extract useful information from multiple scales. Accuracy, recall, Dice coefficient, accuracy and sensitivity were used to evaluate the ability of the model to detect and segment the automatic retinal exudation features of diabetic patients in color fundus images. Results After applying this method, the accuracy, recall, dice coefficient, accuracy and sensitivity of the improved model on the IDRID dataset reached 81.56%, 99.54%, 69.32%, 65.36% and 78.33%, respectively. Compared with the original model, the accuracy and Dice index of the improved model are increased by 2.35% , 3.35% respectively. Conclusion The segmentation method based on U-shaped network can automatically detect and segment the retinal exudation features of fundus images of diabetic patients, which is of great significance for assisting doctors to diagnose diseases more accurately.
Gait recognition is a new technology in biometric recognition and medical treatment which has advantages such as long-distance and non-invasiveness. Depending on the differences between different people's walking postures, we can recognize individuals by characteristics extracted from the images of walking movement. A complete gait recognition process usually includes gait sequence acquisition, gait detection, feature extracting and recognition. In this paper, the commonly used methods of these four processes are introduced, and feature extraction methods are classified from different multi-angle views. And then the new algorithm of multi-view emerged in recent years is highlighted. In addition, this paper summarizes the existing difficulties of gait recognition, and looks into the future development trends of it.
Existing arrhythmia classification methods usually use manual selection of electrocardiogram (ECG) signal features, so that the feature selection is subjective, and the feature extraction is complex, leaving the classification accuracy usually affected. Based on this situation, a new method of arrhythmia automatic classification based on discriminative deep belief networks (DDBNs) is proposed. The morphological features of heart beat signals are automatically extracted from the constructed generative restricted Boltzmann machine (GRBM), then the discriminative restricted Boltzmann machine (DRBM) with feature learning and classification ability is introduced, and arrhythmia classification is performed according to the extracted morphological features and RR interval features. In order to further improve the classification performance of DDBNs, DDBNs are converted to deep neural network (DNN) using the Softmax regression layer for supervised classification in this paper, and the network is fine-tuned by backpropagation. Finally, the Massachusetts Institute of Technology and Beth Israel Hospital Arrhythmia Database (MIT-BIH AR) is used for experimental verification. For training sets and test sets with consistent data sources, the overall classification accuracy of the method is up to 99.84% ± 0.04%. For training sets and test sets with inconsistent data sources, a small number of training sets are extended by the active learning (AL) method, and the overall classification accuracy of the method is up to 99.31% ± 0.23%. The experimental results show the effectiveness of the method in arrhythmia automatic feature extraction and classification. It provides a new solution for the automatic extraction of ECG signal features and classification for deep learning.
Cardiotocography (CTG) is a commonly used technique of electronic fetal monitoring (EFM) for evaluating fetal well-being, which has the disadvantage of lower diagnostic rate caused by subjective factors. To reduce the rate of misdiagnosis and assist obstetricians in making accurate medical decisions, this paper proposed an intelligent assessment approach for analyzing fetal state based on fetal heart rate (FHR) signals. First, the FHR signals from the public database of the Czech Technical University-University Hospital in Brno (CTU-UHB) was preprocessed, and the comprehensive features were extracted. Then the optimal feature subset based on the k-nearest neighbor (KNN) genetic algorithm (GA) was selected. At last the classification using least square support vector machine (LS-SVM) was executed. The experimental results showed that the classification of fetal state achieved better performance using the proposed method in this paper: the accuracy is 91%, sensitivity is 89%, specificity is 94%, quality index is 92%, and area under the receiver operating characteristic curve is 92%, which can assist clinicians in assessing fetal state effectively.
The automatic detection of arrhythmia is of great significance for the early prevention and diagnosis of cardiovascular diseases. Traditional arrhythmia diagnosis is limited by expert knowledge and complex algorithms, and lacks multi-dimensional feature representation capabilities, which is not suitable for wearable electrocardiogram (ECG) monitoring equipment. This study proposed a feature extraction method based on autoregressive moving average (ARMA) model fitting. Different types of heartbeats were used as model inputs, and the characteristic of fast and smooth signal was used to select the appropriate order for the arrhythmia signal to perform coefficient fitting, and complete the ECG feature extraction. The feature vectors were input to the support vector machine (SVM) classifier and K-nearest neighbor classifier (KNN) for automatic ECG classification. MIT-BIH arrhythmia database and MIT-BIH atrial fibrillation database were used to verify in the experiment. The experimental results showed that the feature engineering composed of the fitting coefficients of the ARMA model combined with the SVM classifier obtained a recall rate of 98.2% and a precision rate of 98.4%, and the F1 index was 98.3%. The algorithm has high performance, meets the needs of clinical diagnosis, and has low algorithm complexity. It can use low-power embedded processors for real-time calculations, and it’s suitable for real-time warning of wearable ECG monitoring equipment.
