Emotion is the human brain reacting to objective things. In real life, human emotions are complex and changeable, so research into emotion recognition is of great significance in real life applications. Recently, many deep learning and machine learning methods have been widely applied in emotion recognition based on EEG signals. However, the traditional machine learning method has a major disadvantage in that the feature extraction process is usually cumbersome, which relies heavily on human experts. Then, end-to-end deep learning methods emerged as an effective method to address this disadvantage with the help of raw signal features and time-frequency spectrums. Here, we investigated the application of several deep learning models to the research field of EEG-based emotion recognition, including deep neural networks (DNN), convolutional neural networks (CNN), long short-term memory (LSTM), and a hybrid model of CNN and LSTM (CNN-LSTM). The experiments were carried on the well-known DEAP dataset. Experimental results show that the CNN and CNN-LSTM models had high classification performance in EEG-based emotion recognition, and their accurate extraction rate of RAW data reached 90.12 and 94.17%, respectively. The performance of the DNN model was not as accurate as other models, but the training speed was fast. The LSTM model was not as stable as the CNN and CNN-LSTM models. Moreover, with the same number of parameters, the training speed of the LSTM was much slower and it was difficult to achieve convergence. Additional parameter comparison experiments with other models, including epoch, learning rate, and dropout probability, were also conducted in the paper. Comparison results prove that the DNN model converged to optimal with fewer epochs and a higher learning rate. In contrast, the CNN model needed more epochs to learn. As for dropout probability, reducing the parameters by ~50% each time was appropriate.

Chest X-ray images are widely used in clinical practice such as diagnosis and treatment. The automatic radiology report generation system can effectively reduce the rate of misdiagnosis and missed diagnosis. Previous studies were focused on the long text generation problem of image paragraph, ignoring the characteristics of the image and the auxiliary role of patient background information for diagnosis. In this paper, we propose a new hierarchical model with multi-attention considering the background information. The multi-attention mechanism can focus on the image’s channel and spatial information simultaneously, and map it to the sentence topic. The patient’s background information will be encoded by the neural network first, then it will be aggregated into a vector representation by a multi-layer perception and added to the pre-trained vanilla word embedding, which finally forms a new word embedding after fusion. Our experimental results demonstrated that the model outperforms all baselines, achieving the state-of-the-art performance in terms of accuracy.

A motor imagery (MI) brain-computer interface (BCI) plays an important role in the neurological rehabilitation training for stroke patients. Electroencephalogram (EEG)-based MI BCI has high temporal resolution, which is convenient for real-time BCI control. Therefore, we focus on EEG-based MI BCI in this paper. The identification of MI EEG signals is always quite challenging. Due to high inter-session/subject variability, each subject should spend long and tedious calibration time in collecting amounts of labeled samples for a subject-specific model. To cope with this problem, we present a supervised selective cross-subject transfer learning (sSCSTL) approach which simultaneously makes use of the labeled samples from target and source subjects based on Riemannian tangent space. Since the covariance matrices representing the multi-channel EEG signals belong to the smooth Riemannian manifold, we perform the Riemannian alignment to make the covariance matrices from different subjects close to each other. Then, all aligned covariance matrices are converted into the Riemannian tangent space features to train a classifier in the Euclidean space. To investigate the role of unlabeled samples, we further propose semi-supervised and unsupervised versions which utilize the total samples and unlabeled samples from target subject, respectively. Sequential forward floating search (SFFS) method is executed for source selection. All our proposed algorithms transfer the labeled samples from most suitable source subjects into the feature space of target subject. Experimental results on two publicly available MI datasets demonstrated that our algorithms outperformed several state-of-the-art algorithms using small number of the labeled samples from target subject, especially for good target subjects.

Computer-aided diagnosis (CAD) is an important work which can improve the working efficiency of physicians. With the availability of large-scale data sets, several methods have been proposed to classify pathology on chest X-ray images. However, most methods report performance based on a frontal chest radiograph, ignoring the effect of the lateral chest radiography on the diagnosis. This paper puts forward a kind of model, Dual-Ray Net, of a deep convolutional neural network which can deal with the front and lateral chest radiography at the same time by referring the method of using lateral chest radiography to assist diagnose during the diagnosis used by radiologists. Firstly, we evaluated the performance of parameter migration to small data after pre-training for large datasets. The data sets for pre-training are chest X-ray 14 and ImageNet respectively. The results showed that pre-training with chest X-ray 14 performed better than with the generic dataset ImageNet. Secondly, We evaluated the performance of the Frontal and lateral chest radiographs in different modes of input model for the diagnosis of assisted chest disease. Finally, by comparing different feature fusion methods of addition and concatenation, we found that the fusion effect of concatenation is better, which average AUC reached 0.778. The comparison results show that whether it is a public or a non-public dataset, our Dual-Ray Net (concatenation) architecture shows improved performance in recognizing findings in CXR images when compared to applying separate baseline frontal and lateral classes.

The chest X-ray is a simple and economical medical aid for auxiliary diagnosis, therefore has becomes a routine item for residents’ physical examinations. Based on 40,167 images of chest radiographs and corresponding reports, we explore the abnormality classification problem of chest X-rays by taking the advantage of deep learning techniques. First of all, since the radiology reports are generally templatized by the aberrant physical regions, we propose an annotation method according to the abnormal part in the images. Secondly, building on a small number of reports which are manually annotated by professional radiologists, we employ the Long-Short Term Memory (LSTM) model to automatically annotate the remaining unlabeled data. The result shows that the precision value reaches 0.88 in accurately annotating images, the recall value reaches 0.85, and the F1-score reaches 0.86. Finally, we classify the abnormality in chest X-rays by training convolutional neural networks, and the results show that the average AUC value reaches 0.835.