Search

Pneumoconiosis ranks first among the newly-emerged occupational diseases reported annually in China, and imaging diagnosis is still one of the main clinical diagnostic methods. However, manual reading of films requires high level of doctors, and it is difficult to discriminate the staged diagnosis of pneumoconiosis imaging, and due to the influence of uneven distribution of medical resources and other factors, it is easy to lead to misdiagnosis and omission of diagnosis in primary healthcare institutions. Computer-aided diagnosis system can realize rapid screening of pneumoconiosis in order to assist clinicians in identification and diagnosis, and improve diagnostic efficacy. As an important branch of deep learning, convolutional neural network (CNN) is good at dealing with various visual tasks such as image segmentation, image classification, target detection and so on because of its characteristics of local association and weight sharing, and has been widely used in the field of computer-aided diagnosis of pneumoconiosis in recent years. This paper was categorized into three parts according to the main applications of CNNs (VGG, U-Net, ResNet, DenseNet, CheXNet, Inception-V3, and ShuffleNet) in the imaging diagnosis of pneumoconiosis, including CNNs in pneumoconiosis screening diagnosis, CNNs in staging diagnosis of pneumoconiosis, and CNNs in segmentation of pneumoconiosis foci to conduct a literature review. It aims to summarize the methods, advantages and disadvantages, and optimization ideas of CNN applied to the images of pneumoconiosis, and to provide a reference for the research direction of further development of computer-aided diagnosis of pneumoconiosis.

Computed tomography (CT) imaging is a vital tool for the diagnosis and assessment of lung adenocarcinoma, and using CT images to predict the recurrence-free survival (RFS) of lung adenocarcinoma patients post-surgery is of paramount importance in tailoring postoperative treatment plans. Addressing the challenging task of accurate RFS prediction using CT images, this paper introduces an innovative approach based on self-supervised pre-training and multi-task learning. We employed a self-supervised learning strategy known as "image transformation to image restoration" to pretrain a 3D-UNet network on publicly available lung CT datasets to extract generic visual features from lung images. Subsequently, we enhanced the network's feature extraction capability through multi-task learning involving segmentation and classification tasks, guiding the network to extract image features relevant to RFS. Additionally, we designed a multi-scale feature aggregation module to comprehensively amalgamate multi-scale image features, and ultimately predicted the RFS risk score for lung adenocarcinoma with the aid of a feed-forward neural network. The predictive performance of the proposed method was assessed by ten-fold cross-validation. The results showed that the consistency index (C-index) of the proposed method for predicting RFS and the area under curve (AUC) for predicting whether recurrence occurs within three years reached 0.691 ± 0.076 and 0.707 ± 0.082, respectively, and the predictive performance was superior to that of existing methods. This study confirms that the proposed method has the potential of RFS prediction in lung adenocarcinoma patients, which is expected to provide a reliable basis for the development of individualized treatment plans.

The deep learning-based automatic detection of epilepsy electroencephalogram (EEG), which can avoid the artificial influence, has attracted much attention, and its effectiveness mainly depends on the deep neural network model. In this paper, an attention-based multi-scale residual network (AMSRN) was proposed in consideration of the multiscale, spatio-temporal characteristics of epilepsy EEG and the information flow among channels, and it was combined with multiscale principal component analysis (MSPCA) to realize the automatic epilepsy detection. Firstly, MSPCA was used for noise reduction and feature enhancement of original epilepsy EEG. Then, we designed the structure and parameters of AMSRN. Among them, the attention module (AM), multiscale convolutional module (MCM), spatio-temporal feature extraction module (STFEM) and classification module (CM) were applied successively to signal reexpression with attention weighted mechanism as well as extraction, fusion and classification for multiscale and spatio-temporal features. Based on the Children's Hospital Boston-Massachusetts Institute of Technology ( CHB-MIT) public dataset, the AMSRN model achieved good results in sensitivity (98.56%), F1 score (98.35%), accuracy (98.41%) and precision (98.43%). The results show that AMSRN can make good use of brain network information flow caused by seizures to enhance the difference among channels, and effectively capture the multiscale and spatio-temporal features of EEG to improve the performance of epilepsy detection.

The pace of modern life is accelerating, the pressure of life is gradually increasing, and the long-term accumulation of mental fatigue poses a threat to health. By analyzing physiological signals and parameters, this paper proposes a method that can identify the state of mental fatigue, which helps to maintain a healthy life. The method proposed in this paper is a new recognition method of psychological fatigue state of electrocardiogram signals based on convolutional neural network and long short-term memory. Firstly, the convolution layer of one-dimensional convolutional neural network model is used to extract local features, the key information is extracted through pooling layer, and some redundant data is removed. Then, the extracted features are used as input to the long short-term memory model to further fuse the ECG features. Finally, by integrating the key information through the full connection layer, the accurate recognition of mental fatigue state is successfully realized. The results show that compared with traditional machine learning algorithms, the proposed method significantly improves the accuracy of mental fatigue recognition to 96.3%, which provides a reliable basis for the early warning and evaluation of mental fatigue.

The recurrent neural network architecture improves the processing ability of time-series data. However, issues such as exploding gradients and poor feature extraction limit its application in the automatic diagnosis of mild cognitive impairment (MCI). This paper proposed a research approach for building an MCI diagnostic model using a Bayesian-optimized bidirectional long short-term memory network (BO-BiLSTM) to address this problem. The diagnostic model was based on a Bayesian algorithm and combined prior distribution and posterior probability results to optimize the BO-BiLSTM network hyperparameters. It also used multiple feature quantities that fully reflected the cognitive state of the MCI brain, such as power spectral density, fuzzy entropy, and multifractal spectrum, as the input of the diagnostic model to achieve automatic MCI diagnosis. The results showed that the feature-fused Bayesian-optimized BiLSTM network model achieved an MCI diagnostic accuracy of 98.64% and effectively completed the diagnostic assessment of MCI. In conclusion, based on this optimization, the long short-term neural network model has achieved automatic diagnostic assessment of MCI, providing a new diagnostic model for intelligent diagnosis of MCI.

Aiming at the problems of missing important features, inconspicuous details and unclear textures in the fusion of multimodal medical images, this paper proposes a method of computed tomography (CT) image and magnetic resonance imaging (MRI) image fusion using generative adversarial network (GAN) and convolutional neural network (CNN) under image enhancement. The generator aimed at high-frequency feature images and used double discriminators to target the fusion images after inverse transform; Then high-frequency feature images were fused by trained GAN model, and low-frequency feature images were fused by CNN pre-training model based on transfer learning. Experimental results showed that, compared with the current advanced fusion algorithm, the proposed method had more abundant texture details and clearer contour edge information in subjective representation. In the evaluation of objective indicators, Q AB/F , information entropy (IE), spatial frequency (SF), structural similarity (SSIM), mutual information (MI) and visual information fidelity for fusion (VIFF) were 2.0%, 6.3%, 7.0%, 5.5%, 9.0% and 3.3% higher than the best test results, respectively. The fused image can be effectively applied to medical diagnosis to further improve the diagnostic efficiency.

Fetal electrocardiogram (ECG) signals provide important clinical information for early diagnosis and intervention of fetal abnormalities. In this paper, we propose a new method for fetal ECG signal extraction and analysis. Firstly, an improved fast independent component analysis method and singular value decomposition algorithm are combined to extract high-quality fetal ECG signals and solve the waveform missing problem. Secondly, a novel convolutional neural network model is applied to identify the QRS complex waves of fetal ECG signals and effectively solve the waveform overlap problem. Finally, high quality extraction of fetal ECG signals and intelligent recognition of fetal QRS complex waves are achieved. The method proposed in this paper was validated with the data from the PhysioNet computing in cardiology challenge 2013 database of the Complex Physiological Signals Research Resource Network. The results show that the average sensitivity and positive prediction values of the extraction algorithm are 98.21% and 99.52%, respectively, and the average sensitivity and positive prediction values of the QRS complex waves recognition algorithm are 94.14% and 95.80%, respectively, which are better than those of other research results. In conclusion, the algorithm and model proposed in this paper have some practical significance and may provide a theoretical basis for clinical medical decision making in the future.

Intelligent medical image segmentation methods have been rapidly developed and applied, while a significant challenge is domain shift. That is, the segmentation performance degrades due to distribution differences between the source domain and the target domain. This paper proposed an unsupervised end-to-end domain adaptation medical image segmentation method based on the generative adversarial network (GAN). A network training and adjustment model was designed, including segmentation and discriminant networks. In the segmentation network, the residual module was used as the basic module to increase feature reusability and reduce model optimization difficulty. Further, it learned cross-domain features at the image feature level with the help of the discriminant network and a combination of segmentation loss with adversarial loss. The discriminant network took the convolutional neural network and used the labels from the source domain, to distinguish whether the segmentation result of the generated network is from the source domain or the target domain. The whole training process was unsupervised. The proposed method was tested with experiments on a public dataset of knee magnetic resonance (MR) images and the clinical dataset from our cooperative hospital. With our method, the mean Dice similarity coefficient (DSC) of segmentation results increased by 2.52% and 6.10% to the classical feature level and image level domain adaptive method. The proposed method effectively improves the domain adaptive ability of the segmentation method, significantly improves the segmentation accuracy of the tibia and femur, and can better solve the domain transfer problem in MR image segmentation.

Aiming at the problem that the unbalanced distribution of data in sleep electroencephalogram(EEG) signals and poor comfort in the process of polysomnography information collection will reduce the model's classification ability, this paper proposed a sleep state recognition method using single-channel EEG signals (WKCNN-LSTM) based on one-dimensional width kernel convolutional neural networks(WKCNN) and long-short-term memory networks (LSTM). Firstly, the wavelet denoising and synthetic minority over-sampling technique-Tomek link (SMOTE-Tomek) algorithm were used to preprocess the original sleep EEG signals. Secondly, one-dimensional sleep EEG signals were used as the input of the model, and WKCNN was used to extract frequency-domain features and suppress high-frequency noise. Then, the LSTM layer was used to learn the time-domain features. Finally, normalized exponential function was used on the full connection layer to realize sleep state. The experimental results showed that the classification accuracy of the one-dimensional WKCNN-LSTM model was 91.80% in this paper, which was better than that of similar studies in recent years, and the model had good generalization ability. This study improved classification accuracy of single-channel sleep EEG signals that can be easily utilized in portable sleep monitoring devices.

In deep learning-based image registration, the deformable region with complex anatomical structures is an important factor affecting the accuracy of network registration. However, it is difficult for existing methods to pay attention to complex anatomical regions of images. At the same time, the receptive field of the convolutional neural network is limited by the size of its convolution kernel, and it is difficult to learn the relationship between the voxels with far spatial location, making it difficult to deal with the large region deformation problem. Aiming at the above two problems, this paper proposes a cascaded multi-level registration network model based on transformer, and equipped it with a difficult deformable region perceptron based on mean square error. The difficult deformation perceptron uses sliding window and floating window techniques to retrieve the registered images, obtain the difficult deformation coefficient of each voxel, and identify the regions with the worst registration effect. In this study, the cascaded multi-level registration network model adopts the difficult deformation perceptron for hierarchical connection, and the self-attention mechanism is used to extract global features in the basic registration network to optimize the registration results of different scales. The experimental results show that the method proposed in this paper can perform progressive registration of complex deformation regions, thereby optimizing the registration results of brain medical images, which has a good auxiliary effect on the clinical diagnosis of doctors.

[ Abstract ]Automatic and accurate segmentation of lung parenchyma is essential for assisted diagnosis of lung cancer. In recent years, researchers in the field of deep learning have proposed a number of improved lung parenchyma segmentation methods based on U-Net. However, the existing segmentation methods ignore the complementary fusion of semantic information in the feature map between different layers and fail to distinguish the importance of different spaces and channels in the feature map. To solve this problem, this paper proposes the double scale parallel attention (DSPA) network (DSPA-Net) architecture, and introduces the DSPA module and the atrous spatial pyramid pooling (ASPP) module in the "encoder-decoder" structure. Among them, the DSPA module aggregates the semantic information of feature maps of different levels while obtaining accurate space and channel information of feature map with the help of cooperative attention (CA). The ASPP module uses multiple parallel convolution kernels with different void rates to obtain feature maps containing multi-scale information under different receptive fields. The two modules address multi-scale information processing in feature maps of different levels and in feature maps of the same level, respectively. We conducted experimental verification on the Kaggle competition dataset. The experimental results prove that the network architecture has obvious advantages compared with the current mainstream segmentation network. The values of dice similarity coefficient (DSC) and intersection on union (IoU) reached 0.972 ± 0.002 and 0.945 ± 0.004, respectively. This paper achieves automatic and accurate segmentation of lung parenchyma and provides a reference for the application of attentional mechanisms and multi-scale information in the field of lung parenchyma segmentation.

Blood velocity inversion based on magnetoelectric effect is helpful for the development of daily monitoring of vascular stenosis, but the accuracy of blood velocity inversion and imaging resolution still need to be improved. Therefore, a convolutional neural network (CNN) based inversion imaging method for intravascular blood flow velocity was proposed in this paper. Firstly, unsupervised learning CNN is constructed to extract weight matrix representation information to preprocess voltage data. Then the preprocessing results are input to supervised learning CNN, and the blood flow velocity value is output by nonlinear mapping. Finally, angiographic images are obtained. In this paper, the validity of the proposed method is verified by constructing data set. The results show that the correlation coefficients of blood velocity inversion in vessel location and stenosis test are 0.884 4 and 0.972 1, respectively. The above research shows that the proposed method can effectively reduce the information loss during the inversion process and improve the inversion accuracy and imaging resolution, which is expected to assist clinical diagnosis.

Accurate segmentation of ground glass nodule (GGN) is important in clinical. But it is a tough work to segment the GGN, as the GGN in the computed tomography images show blur boundary, irregular shape, and uneven intensity. This paper aims to segment GGN by proposing a fully convolutional residual network, i.e. , residual network based on atrous spatial pyramid pooling structure and attention mechanism (ResAANet). The network uses atrous spatial pyramid pooling (ASPP) structure to expand the feature map receptive field and extract more sufficient features, and utilizes attention mechanism, residual connection, long skip connection to fully retain sensitive features, which is extracted by the convolutional layer. First, we employ 565 GGN provided by Shanghai Chest Hospital to train and validate ResAANet, so as to obtain a stable model. Then, two groups of data selected from clinical examinations (84 GGN) and lung image database consortium (LIDC) dataset (145 GGN) were employed to validate and evaluate the performance of the proposed method. Finally, we apply the best threshold method to remove false positive regions and obtain optimized results. The average dice similarity coefficient (DSC) of the proposed algorithm on the clinical dataset and LIDC dataset reached 83.46%, 83.26% respectively, the average Jaccard index (IoU) reached 72.39%, 71.56% respectively, and the speed of segmentation reached 0.1 seconds per image. Comparing with other reported methods, our new method could segment GGN accurately, quickly and robustly. It could provide doctors with important information such as nodule size or density, which assist doctors in subsequent diagnosis and treatment.

Electrocardiogram (ECG) can visually reflect the physiological electrical activity of human heart, which is important in the field of arrhythmia detection and classification. To address the negative effect of label imbalance in ECG data on arrhythmia classification, this paper proposes a nested long short-term memory network (NLSTM) model for unbalanced ECG signal classification. The NLSTM is built to learn and memorize the temporal characteristics in complex signals, and the focal loss function is used to reduce the weights of easily identifiable samples. Then the residual attention mechanism is used to modify the assigned weights according to the importance of sample characteristic to solve the sample imbalance problem. Then the synthetic minority over-sampling technique is used to perform a simple manual oversampling process on the Massachusetts institute of technology and Beth Israel hospital arrhythmia (MIT-BIH-AR) database to further increase the classification accuracy of the model. Finally, the MIT-BIH arrhythmia database is applied to experimentally verify the above algorithms. The experimental results show that the proposed method can effectively solve the issues of imbalanced samples and unremarkable features in ECG signals, and the overall accuracy of the model reaches 98.34%. It also significantly improves the recognition and classification of minority samples and has provided a new feasible method for ECG-assisted diagnosis, which has practical application significance.

Aiming at the problems of individual differences in the asynchrony process of human lower limbs and random changes in stride during walking, this paper proposes a method for gait recognition and prediction using motion posture signals. The research adopts an optimized gated recurrent unit (GRU) network algorithm based on immune particle swarm optimization (IPSO) to establish a network model that takes human body posture change data as the input, and the posture change data and accuracy of the next stage as the output, to realize the prediction of human body posture changes. This paper first clearly outlines the process of IPSO's optimization of the GRU algorithm. It collects human body posture change data of multiple subjects performing flat-land walking, squatting, and sitting leg flexion and extension movements. Then, through comparative analysis of IPSO optimized recurrent neural network (RNN), long short-term memory (LSTM) network, GRU network classification and prediction, the effectiveness of the built model is verified. The test results show that the optimized algorithm can better predict the changes in human posture. Among them, the root mean square error (RMSE) of flat-land walking and squatting can reach the accuracy of 10 -3 , and the RMSE of sitting leg flexion and extension can reach the accuracy of 10 -2 . The R 2 value of various actions can reach above 0.966. The above research results show that the optimized algorithm can be applied to realize human gait movement evaluation and gait trend prediction in rehabilitation treatment, as well as in the design of artificial limbs and lower limb rehabilitation equipment, which provide a reference for future research to improve patients' limb function, activity level, and life independence ability.

Most of the existing near-infrared noninvasive blood glucose detection models focus on the relationship between near-infrared absorbance and blood glucose concentration, but do not consider the impact of human physiological state on blood glucose concentration. In order to improve the performance of prediction model, particle swarm optimization (PSO) algorithm was used to train the structure paramters of back propagation (BP) neural network. Moreover, systolic blood pressure, pulse rate, body temperature and 1 550 nm absorbance were introduced as input variables of blood glucose concentration prediction model, and BP neural network was used as prediction model. In order to solve the problem that traditional BP neural network is easy to fall into local optimization, a hybrid model based on PSO-BP was introduced in this paper. The results showed that the prediction effect of PSO-BP model was better than that of traditional BP neural network. The prediction root mean square error and correlation coefficient of ten-fold cross-validation were 0.95 mmol/L and 0.74, respectively. The Clarke error grid analysis results showed that the proportion of model prediction results falling into region A was 84.39%, and the proportion falling into region B was 15.61%, which met the clinical requirements. The model can quickly measure the blood glucose concentration of the subject, and has relatively high accuracy.

As one of the non-invasive imaging techniques, myocardial perfusion imaging provides a basis for the diagnosis of myocardial ischemia in coronary heart disease. Aiming at the bull-eye image in myocardial perfusion imaging, this paper proposed a branching structure, which included multi-layer transposed convolution up-sampling concatenate module and four-channel weighted channels attention module, and the output results of the branch structure were fused with the output results of trunk U-Net, to achieve accurate segmentation of the cardiac ischemia missing degree in myocardial perfusion bull-eye image. The experimental results show that the multi-layer transposed convolution up-sampling concatenate module realizes the fusion of different depth feature maps, and effectively reduces the interference of the severe sparse degree which is similar to the missing degree on the segmentation. Four-channel weighted attention module can further improve the ability to distinguish between the two similar degrees and the ability to learn edge details of the targets, and retain more abundant edge details features. The experimental data came from Tianjin Medical University General Hospital, Tianjin TEDA Hospital, Tianjin First Central Hospital and Third Central Hospital. The Jaccard scores in the self-built dataset was 5.00% higher than that of U-Net. The model presented in this paper is superior to other optimized models based on U-Net, and the subjective evaluation meets the accuracy requirements for clinical diagnosis.

Otitis media is one of the common ear diseases, and its accurate diagnosis can prevent the deterioration of conductive hearing loss and avoid the overuse of antibiotics. At present, the diagnosis of otitis media mainly relies on the doctor's visual inspection based on the images fed back by the otoscope equipment. Due to the quality of otoscope equipment pictures and the doctor's diagnosis experience, this subjective examination has a relatively high rate of misdiagnosis. In response to this problem, this paper proposes the use of faster region convolutional neural networks to analyze clinically collected digital otoscope pictures. First, through image data enhancement and preprocessing, the number of samples in the clinical otoscope dataset was expanded. Then, according to the characteristics of the otoscope picture, the convolutional neural network was selected for feature extraction, and the feature pyramid network was added for multi-scale feature extraction to enhance the detection ability. Finally, a faster region convolutional neural network with anchor size optimization and hyperparameter adjustment was used for identification, and the effectiveness of the method was tested through a randomly selected test set. The results showed that the overall recognition accuracy of otoscope pictures in the test samples reached 91.43%. The above studies show that the proposed method effectively improves the accuracy of otoscope picture classification, and is expected to assist clinical diagnosis.

General anesthesia is an essential part of surgery to ensure the safety of patients. Electroencephalogram (EEG) has been widely used in anesthesia depth monitoring for abundant information and the ability of reflecting the brain activity. The paper proposes a method which combines wavelet transform and artificial neural network (ANN) to assess the depth of anesthesia. Discrete wavelet transform was used to decompose the EEG signal, and the approximation coefficients and detail coefficients were used to calculate the 9 characteristic parameters. Kruskal-Wallis statistical test was made to these characteristic parameters, and the test showed that the parameters were statistically significant for the differences of the four levels of anesthesia: awake, light anesthesia, moderate anesthesia and deep anesthesia ( P < 0.001). The 9 characteristic parameters were used as the input of ANN, the bispectral index (BIS) was used as the reference output, and the method was evaluated by the data of 8 patients during general anesthesia. The accuracy of the method in the classification of the four anesthesia levels of the test set in the 7:3 set-out method was 85.98%, and the correlation coefficient with the BIS was 0.977 0. The results show that this method can better distinguish four different anesthesia levels and has broad application prospects for monitoring the depth of anesthesia.

Under the current situation of the rapid development of brain-like artificial intelligence and the increasingly complex electromagnetic environment, the most bionic and anti-interference spiking neural network has shown great potential in computing speed, real-time information processing, and spatiotemporal data processing. Spiking neural network is the core part of brain-like artificial intelligence, which realizes brain-like computing by simulating the structure of biological neural network and the way of information transmission. This article first summarizes the advantages and disadvantages of the five models, and analyzes the characteristics of several network topologies. Then, it summarizes the spiking neural network algorithms. The unsupervised learning based on spike timing dependent plasticity (STDP) rules and four types of supervised learning algorithms are analyzed. Finally, the research on brain-like neuromorphic chips at home and abroad are reviewed. This paper aims to provide learning ideas and research directions for new colleagues in the field of spiking neural network.

The background of abdominal computed tomography (CT) images is complex, and kidney tumors have different shapes, sizes and unclear edges. Consequently, the segmentation methods applying to the whole CT images are often unable to effectively segment the kidney tumors. To solve these problems, this paper proposes a multi-scale network based on cascaded 3D U-Net and DeepLabV3+ for kidney tumor segmentation, which uses atrous convolution feature pyramid to adaptively control receptive field. Through the fusion of high-level and low-level features, the segmented edges of large tumors and the segmentation accuracies of small tumors are effectively improved. A total of 210 CT data published by Kits2019 were used for five-fold cross validation, and 30 CT volume data collected from Suzhou Science and Technology Town Hospital were independently tested by trained segmentation models. The results of five-fold cross validation experiments showed that the Dice coefficient, sensitivity and precision were 0.796 2 ± 0.274 1, 0.824 5 ± 0.276 3, and 0.805 1 ± 0.284 0, respectively. On the external test set, the Dice coefficient, sensitivity and precision were 0.817 2 ± 0.110 0, 0.829 6 ± 0.150 7, and 0.831 8 ± 0.116 8, respectively. The results show a great improvement in the segmentation accuracy compared with other semantic segmentation methods.

Emotion plays an important role in people's cognition and communication. By analyzing electroencephalogram (EEG) signals to identify internal emotions and feedback emotional information in an active or passive way, affective brain-computer interactions can effectively promote human-computer interaction. This paper focuses on emotion recognition using EEG. We systematically evaluate the performance of state-of-the-art feature extraction and classification methods with a public-available dataset for emotion analysis using physiological signals (DEAP). The common random split method will lead to high correlation between training and testing samples. Thus, we use block-wise K fold cross validation. Moreover, we compare the accuracy of emotion recognition with different time window length. The experimental results indicate that 4 s time window is appropriate for sampling. Filter-bank long short-term memory networks (FBLSTM) using differential entropy features as input was proposed. The average accuracy of low and high in valance dimension, arousal dimension and combination of the four in valance-arousal plane is 78.8%, 78.4% and 70.3%, respectively. These results demonstrate the advantage of our emotion recognition model over the current studies in terms of classification accuracy. Our model might provide a novel method for emotion recognition in affective brain-computer interactions.

Human chromosomes karyotyping is an important means to diagnose genetic diseases. Chromosome image type recognition is a key step in the karyotyping process. Accurate and efficient identification is of great significance for automatic chromosome karyotyping. In this paper, we propose a model named segmentally recalibrated dense convolutional network (SR-DenseNet). In each stage of the model, the dense connected network layers is used to extract the features of different abstract levels of chromosomes automatically, and then the concatenation of all the layers which extract different local features is recalibrated with squeeze-and-excitation (SE) block. SE blocks explicitly construct learnable structures for importance of the features. Then a model fusion method is proposed and an expert group of chromosome recognition models is constructed. On the public available Copenhagen chromosome recognition dataset (G-bands) the proposed model achieves error rate of only 1.60%, and with model fusion the error further drops to 0.99%. On the Padova chromosome dataset (Q-bands) the model gets the corresponding error rate of 6.67%, and with model fusion the error further drops to 5.98%. The experimental results show that the method proposed in this paper is effective and has the potential to realize the automation of chromosome type recognition.

With the rapid improvement of the perception and computing capacity of mobile devices such as smart phones, human activity recognition using mobile devices as the carrier has been a new research hot-spot. The inertial information collected by the acceleration sensor in the smart mobile device is used for human activity recognition. Compared with the common computer vision recognition, it has the following advantages: convenience, low cost, and better reflection of the essence of human motion. Based on the WISDM data set collected by smart phones, the inertial navigation information and the deep learning algorithm-convolutional neural network (CNN) were adopted to build a human activity recognition model in this paper. The K nearest neighbor algorithm (KNN) and the random forest algorithm were compared with the CNN network in the recognition accuracy to evaluate the performance of the CNN network. The classification accuracy of CNN model reached 92.73%, which was much higher than KNN and random forest. Experimental results show that the CNN algorithm model can achieve more accurate human activity recognition and has broad application prospects in predicting and promoting human health.

Compared with the previous automatic segmentation neural network for the target area which considered the target area as an independent area, a stacked neural network which uses the position and shape information of the organs around the target area to regulate the shape and position of the target area through the superposition of multiple networks and fusion of spatial position information to improve the segmentation accuracy on medical images was proposed in this paper. Taking the Graves' ophthalmopathy disease as an example, the left and right radiotherapy target areas were segmented by the stacked neural network based on the fully convolutional neural network. The volume Dice similarity coefficient (DSC) and bidirectional Hausdorff distance (HD) were calculated based on the target area manually drawn by the doctor. Compared with the full convolutional neural network, the stacked neural network segmentation results can increase the volume DSC on the left and right sides by 1.7% and 3.4% respectively, while the two-way HD on the left and right sides decrease by 0.6. The results show that the stacked neural network improves the degree of coincidence between the automatic segmentation result and the doctor's delineation of the target area, while reducing the segmentation error of small areas. The stacked neural network can effectively improve the accuracy of the automatic delineation of the radiotherapy target area of Graves' ophthalmopathy.

Inferior myocardial infarction is an acute ischemic heart disease with high mortality, which is easy to induce life-threatening complications such as arrhythmia, heart failure and cardiogenic shock. Therefore, it is of great clinical value to carry out accurate and efficient early diagnosis of inferior myocardial infarction. Electrocardiogram is the most sensitive means for early diagnosis of inferior myocardial infarction. This paper proposes a method for detecting inferior myocardial infarction based on densely connected convolutional neural network. The method uses the original electrocardiogram (ECG) signals of serially connected Ⅱ, Ⅲ and aVF leads as the input of the model and extracts the robust features of the ECG signals by using the scale invariance of the convolutional layers. The characteristic transmission of ECG signals is enhanced by the dense connectivity between different layers, so that the network can automatically learn the effective features with strong robustness and high recognition, so as to achieve accurate detection of inferior myocardial infarction. The Physikalisch Technische Bundesanstalt diagnosis public ECG database was used for verification. The accuracy, sensitivity and specificity of the model reached 99.95%, 100% and 99.90%, respectively. The accuracy, sensitivity and specificity of the model are also over 99% even though the noise exists. Based on the results of this study, it is expected that the method can be introduced in the clinical environment to help doctors quickly diagnose inferior myocardial infarction in the future.

Alzheimer's disease (AD) is a typical neurodegenerative disease, which is clinically manifested as amnesia, loss of language ability and self-care ability, and so on. So far, the cause of the disease has still been unclear and the course of the disease is irreversible, and there has been no cure for the disease yet. Hence, early prognosis of AD is important for the development of new drugs and measures to slow the progression of the disease. Mild cognitive impairment (MCI) is a state between AD and healthy controls (HC). Studies have shown that patients with MCI are more likely to develop AD than those without MCI. Therefore, accurate screening of MCI patients has become one of the research hotspots of early prognosis of AD. With the rapid development of neuroimaging techniques and deep learning, more and more researchers employ deep learning methods to analyze brain neuroimaging images, such as magnetic resonance imaging (MRI), for early prognosis of AD. Hence, in this paper, a three-dimensional multi-slice classifiers ensemble based on convolutional neural network (CNN) and ensemble learning for early prognosis of AD has been proposed. Compared with the CNN classification model based on a single slice, the proposed classifiers ensemble based on multiple two-dimensional slices from three dimensions could use more effective information contained in MRI to improve classification accuracy and stability in a parallel computing mode.

The application of minimally invasive surgical tool detection and tracking technology based on deep learning in minimally invasive surgery is currently a research hotspot. This paper firstly expounds the relevant technical content of the minimally invasive surgery tool detection and tracking, which mainly introduces the advantages based on deep learning algorithm. Then, this paper summarizes the algorithm for detection and tracking surgical tools based on fully supervised deep neural network and the emerging algorithm for detection and tracking surgical tools based on weakly supervised deep neural network. Several typical algorithm frameworks and their flow charts based on deep convolutional and recurrent neural networks are summarized emphatically, so as to enable researchers in relevant fields to understand the current research progress more systematically and provide reference for minimally invasive surgeons to select navigation technology. In the end, this paper provides a general direction for the further research of minimally invasive surgical tool detection and tracking technology based on deep learning.

With the development of image-guided surgery and radiotherapy, the demand for medical image registration is stronger and the challenge is greater. In recent years, deep learning, especially deep convolution neural networks, has made excellent achievements in medical image processing, and its research in registration has developed rapidly. In this paper, the research progress of medical image registration based on deep learning at home and abroad is reviewed according to the category of technical methods, which include similarity measurement with an iterative optimization strategy, direct estimation of transform parameters, etc. Then, the challenge of deep learning in medical image registration is analyzed, and the possible solutions and open research are proposed.

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.

The rapid development of artificial intelligence put forward higher requirements for the computational speed, resource consumption and the biological interpretation of computational neuroscience. Spiking neuron networks can carry a large amount of information, and realize the imitation of brain information processing. However, its hardware is an important way to realize its powerful computing ability, and it is also a challenging technical problem. The memristor currently is the electronic devices that functions closest to the neuron synapse, and able to respond to spike voltage in a highly similar spike timing dependent plasticity (STDP) mechanism with a biological brain, and has become a research hotspot to construct spiking neuron networks hardware circuit in recent years. Through consulting the relevant literature at home and abroad, this paper has made a thorough understanding and introduction to the research work of the spiking neuron networks based on the memristor in recent years.

This paper performs a comprehensive study on the computer-aided detection for the medical diagnosis with deep learning. Based on the region convolution neural network and the prior knowledge of target, this algorithm uses the region proposal network, the region of interest pooling strategy, introduces the multi-task loss function: classification loss, bounding box localization loss and object rotation loss, and optimizes it by end-to-end. For medical image it locates the target automatically, and provides the localization result for the next stage task of segmentation. For the detection of left ventricular in echocardiography, proposed additional landmarks such as mitral annulus, endocardial pad and apical position, were used to estimate the left ventricular posture effectively. In order to verify the robustness and effectiveness of the algorithm, the experimental data of ultrasonic and nuclear magnetic resonance images are selected. Experimental results show that the algorithm is fast, accurate and effective.

Multi-layer perceptron (MLP) neural network belongs to multi-layer feedforward neural network, and has the ability and characteristics of high intelligence. It can realize the complex nonlinear mapping by its own learning through the network. Bipolar disorder is a serious mental illness with high recurrence rate, high self-harm rate and high suicide rate. Most of the onset of the bipolar disorder starts with depressive episode, which can be easily misdiagnosed as unipolar depression and lead to a delayed treatment so as to influence the prognosis. The early identifica- tion of bipolar disorder is of great importance for patients with bipolar disorder. Due to the fact that the process of early identification of bipolar disorder is nonlinear, we in this paper discuss the MLP neural network application in early identification of bipolar disorder. This study covered 250 cases, including 143 cases with recurrent depression and 107 cases with bipolar disorder, and clinical features were statistically analyzed between the two groups. A total of 42 variables with significant differences were screened as the input variables of the neural network. Part of the samples were randomly selected as the learning sample, and the other as the test sample. By choosing different neu- ral network structures, all results of the identification of bipolar disorder were relatively good, which showed that MLP neural network could be used in the early identification of bipolar disorder.

A new method based on convolution kernel compensation (CKC) for decomposing multi-channel surface electromyogram (sEMG) signals is proposed in this paper. Unsupervised learning and clustering function of self-organizing map (SOM) neural network are employed in this method. An initial innervations pulse train (IPT) is firstly estimated, some time instants corresponding to the highest peaks from the initial IPT are clustered by SOM neural network. Then the final IPT can be obtained from the observations corresponding to these time instants. In this paper, the proposed method was tested on the simulated signal, the influence of signal to noise ratio (SNR), the number of groups clustered by SOM and the number of highest peaks selected from the initial pulse train on the number of reconstructed sources and the pulse accuracy were studied, and the results show that the proposed approach is effective in decomposing multi-channel sEMG signals.

The glomerular filtration rate (GFR) is an important index of renal function with advantages of accurate, sensitive, stable, repetitive and accurate measurement of renal functions. The GFR is mainly determined in the clear rate of radioisotope markers (99m)Tc-DTPA, the process of which is complicated, long time-taking, high cost, and radioactively injuring. Recently, the methods using other renal function parameters measured easily to create the mathematic models and to estimate GFR are being investigated in the world. In this paper, for the renal function data of the Chinese, the efficiency in the three existing GFR formulas has been optimized with multi-function optimization techniques and the accuracy of the computation has been improved. Then the BP neural network technology is used for establishing a new GFR formula, which is a simpler form and obtains higher precision result than the formulas existed. The rmse and P30 of the new formula are better than those of the old ones.

This paper generalizes the seizures characterized with paroxysmal generalized spike and wave discharges (GSWDs) in the EEG. Recent studies showed that GSWDs disrupt specific neural networks only rather than the entire brain homogenously. Simultaneous EEG and functional MRI (EEG-fMRDI) provides a high spatiotemporal resolution method for uncovering the regions of the brain showing changes in metabolism and blood flow during epileptic activity. Human EEG-fMRI studies to date have revealed the blood oxygenation level dependent (BOLD) signal changes in response to GSWDs in some specific brain regions. Most studies have noted similar BOLD signals decrease in the bilateral cortical regions including frontal, frontal-parietal, posterior cingulated and precuneus cortex, as well as in the basal ganglia, and BOLD signals increase in the bilateral thalamic. Further studies demonstrated that BOLD signals in different regions were dynamic changes in the time course of GSWDs and BOLD changes in the cortical areas occurred before in the thalamus. These cortical-subcortical structures may form the neural networks associated with GSWDs generation and maintenance. More sophisticated analytic techniques will be developed to explore the BOLD time-course of GSWDs and identify the brain structures involved in seizure onset and discharges propagation respectively. The sub-network associated with different behavioral deficits between interical and ictal GSWDs, and the different subtypes of generalized seizures will be further studied. The functional connectivity of the nodes of the neural network of GSWDs can also be further investigated. A better understanding of the neural network responsible for GSWDs generation may help to develop new therapeutic interventions.

This paper is to provide a basis for the establishment of an early diagnostic system for hypoxic-ischemic encephalopathy (HIE) by performing segmentation and feature extraction of lesions on the MR images of neonatal babies with HIE. The segmentation on MR images of HIE based on the genetic algorithm (GA) combined with a pulse-coupled neural network (PCNN) were carried out. There were better segmentation results by using PCNN segmentation based on GA than PCNN segmentation with fixed parameters. The data suggested that a PCNN based on GA could provide effective assistance for diagnosis and research.

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.

In this paper, the ECG de-noising technology based on wavelet neural networks (WNN) is used to deal with the noises in Electrocardiogram (ECG) signal. The structure of WNN, which has the outstanding nonlinear mapping capability, is designed as a nonlinear filter used for ECG to cancel the baseline wander, electromyo-graphical interference and powerline interference. The network training algorithm and de-noising experiments results are presented, and some key points of the WNN filter using ECG de-noising are discussed.

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.

Noninvasive detection of tissue lesion is an important aided way in the process of high intensity focused ultrasound (HIFU) treatment. A new method is proposed for noninvasive detection of tissue lesion based on self-organizing maps (SOM) network. Tissue lesion, created by HIFU exposures, inevitably leads to change in B-mode image, since some new interfaces will appear in tissue. The difference-sub-image near focus of HIFU, where the tissue lesion appears, is used. Feature of lesion in difference-sub-image is extracted using the method based on SOM network. Experimental results show that the performance of the method proposed in this paper is better than that of others, such as the methods based on average gray, energy and standard deviation of difference-sub-image.

An automatic detection of Electrocardiogram (ECG) using Quantum Neural Networks (QNN) with multiresolution analysis is given in the paper. QNN originates from exploiting BP neural networks. With the quantum neurons, QNN can model the levels of uncertainty arising from complex classification problems. Its potential advantage over conventional methods is based on the argument for reliability. The fuzzy feature is expected to enhance the reliability of the network, which is critical for improving desirable diagnosis accuracy. And wavelet transform with multiresolution analysis is adopted in ECG pretreatment for reducing the numbers of neurons. So it can improve convergence speed of the network. The results of simulation confirmed the feasibility of the proposed approach.

The similarity measures between Vague sets are one of the most important technologies in Vague sets, In this paper, the new similarity measures based on Huang Guoshun's related works are presented and applied in electrocardiogram auto-recognition. Based on medical requiresments, in this paper, the characteristic parameters of signals from Massachusettes Institute of Technology (database) have been picked up and studied with BP neural network. In the end, the electrocardiogram samples are classified with the use of those characteristic parameters. The result shows that the accuracy of recognition goes up to 99.04%.

Compared with other sensory system, olfactory neural system may be the most unknown one. And it is reported that the research of the complicated olfactory system is beneficial to clarifying the whole mechanism of the sensory system. Focused on spatiotemporal coding and decoding mechanism, the studies on the olfactory neural system recognition models are especially introduced. Finally, this paper presents the research work carried out in our lab, and prospects the development of this field in the future.

This paper presents a method of using feature searching algorithm based on maximal divergence value to get the optimized feature combinations at different dimensions from feature space. Feature space is obtained through wavelet transform on ECG beat. Then the feature vector is determined by analyzing the changes of divergence value of those optimized feature combinations along with the dimensions. BP artificial neural network is trained by the feature vector and four types of ECG beats(normal beat, left bundle branch block beat, right bundle branch block beat and paced beat) obtained from MIT-BIH database are classified with a success of 93.9%.

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.

Support vector machine (SVM) has shown its excellent learning and generalization ability for the binary classification of real problems and has been extensively employed in many areas. In this paper, SVM, K-Nearest Neighbor, Decision Tree C4.5 and Artificial Neural Network were applied to identify cancer patients and normal individuals using the concentrations of 6 elements including macroelements (Ca, Mg) and microelements (Ba, Cu, Se, Zn) in human blood. It was demonstrated, by using the normalized features instead of the original features, the classification performances can be improved from 91.89% to 95.95%, from 83.78% to 93.24%, and from 90.54% to 94.59% for SVM, K-NN and ANN respectively, whereas that of C4.5 keeps unchangeable. The best average accuracy of SVM with linear dot kernel by using 5-fold cross validation reaches 95.95%, and is superior to those of other classifiers based on K-NN (93.24%), C4.5 (79.73%), and ANN (94.59%). The study suggests that support vector machine is capable of being used as a potential application methodology for SVM-aided clinical cancer diagnosis.

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.

The purpose of this article is to evaluate the role of quantitative margin features in the computer-aided diagnosis of malignant and benign solid breast masses using sonographic imaging. The tumour was seperated by the expert. Three contour features circurity (C), area ratio (A) and length width ratio (LWR) was caculated from the tumour contour. Then back-propagation (BP) neural network with contour features was used to classify tumors into benign and malignant. Results from 119 ultrasonic images have been applied in this experiment. BP neural network yielded the following results: 89.7% and 73.5% respectively. The methods applied in this paper are helpful to raise the correctance of breast cancer diagnosis.

This paper discusses the recognition of heart sound for evaluating the cardiac contractility change trend, which includes heart sound samples recorded at different exercise condition. Especially, focused on the recognition of heart sound recorded after high intensity exercise workload. The algorithm proposed consisted of two correlative methods. The first was to recognize heart sound recorded at rest and after low intensity exercise workloads by probabilistic neural network and the second was to recognize heart sound recorded after high intensity exercise workloads based on the characteristic of heart sound. Both methods have two consecutive phases. Firstly, all peaks, including the peaks of both heart sounds and noise, are marked by a repetitive threshold detecting algorithm. Secondly, probabilistic neural network is employed to classify the peaks detected in the first phase into Si, S2, and noise. Finally, the performance of the algorithm was evaluated using 45 digital heart sound recordings including normal and abnormal heart sound, which were recorded at rest and after low intensity exercise workloads, and 28 digital heart sound recordings recorded after high intensity exercise workloads. The results showed that over 94% of heart sound samples were classified and recognized correctly. Moreover, the reasons for the wrong classification, of which omitting and misdetection are two main problems, are also discussed and solutions are proposed. So this method can be improved and refined in following studies. In conclusion, this algorithm is a reliable approach to detect and classify heart sounds, providing a solid basis for further heart sound analysis.