Objective To develop a deep learning system for CT images to assist in the diagnosis of thoracolumbar fractures and analyze the feasibility of its clinical application. Methods Collected from West China Hospital of Sichuan University from January 2019 to March 2020, a total of 1256 CT images of thoracolumbar fractures were annotated with a unified standard through the Imaging LabelImg system. All CT images were classified according to the AO Spine thoracolumbar spine injury classification. The deep learning system in diagnosing ABC fracture types was optimized using 1039 CT images for training and validation, of which 1004 were used as the training set and 35 as the validation set; the rest 217 CT images were used as the test set to compare the deep learning system with the clinician’s diagnosis. The deep learning system in subtyping A was optimized using 581 CT images for training and validation, of which 556 were used as the training set and 25 as the validation set; the rest 104 CT images were used as the test set to compare the deep learning system with the clinician’s diagnosis. Results The accuracy and Kappa coefficient of the deep learning system in diagnosing ABC fracture types were 89.4% and 0.849 (P<0.001), respectively. The accuracy and Kappa coefficient of subtyping A were 87.5% and 0.817 (P<0.001), respectively. Conclusions The classification accuracy of the deep learning system for thoracolumbar fractures is high. This approach can be used to assist in the intelligent diagnosis of CT images of thoracolumbar fractures and improve the current manual and complex diagnostic process.
Cardiovascular disease (CVD) is one of the leading causes of death worldwide. Heart sound classification plays a key role in the early detection of CVD. The difference between normal and abnormal heart sounds is not obvious. In this paper, in order to improve the accuracy of the heart sound classification model, we propose a heart sound feature extraction method based on bispectral analysis and combine it with convolutional neural network (CNN) to classify heart sounds. The model can effectively suppress Gaussian noise by using bispectral analysis and can effectively extract the features of heart sound signals without relying on the accurate segmentation of heart sound signals. At the same time, the model combines with the strong classification performance of convolutional neural network and finally achieves the accurate classification of heart sound. According to the experimental results, the proposed algorithm achieves 0.910, 0.884 and 0.940 in terms of accuracy, sensitivity and specificity under the same data and experimental conditions, respectively. Compared with other heart sound classification algorithms, the proposed algorithm shows a significant improvement and strong robustness and generalization ability, so it is expected to be applied to the auxiliary detection of congenital heart disease.
The count and recognition of white blood cells in blood smear images play an important role in the diagnosis of blood diseases including leukemia. Traditional manual test results are easily disturbed by many factors. It is necessary to develop an automatic leukocyte analysis system to provide doctors with auxiliary diagnosis, and blood leukocyte segmentation is the basis of automatic analysis. In this paper, we improved the U-Net model and proposed a segmentation algorithm of leukocyte image based on dual path and atrous spatial pyramid pooling. Firstly, the dual path network was introduced into the feature encoder to extract multi-scale leukocyte features, and the atrous spatial pyramid pooling was used to enhance the feature extraction ability of the network. Then the feature decoder composed of convolution and deconvolution was used to restore the segmented target to the original image size to realize the pixel level segmentation of blood leukocytes. Finally, qualitative and quantitative experiments were carried out on three leukocyte data sets to verify the effectiveness of the algorithm. The results showed that compared with other representative algorithms, the proposed blood leukocyte segmentation algorithm had better segmentation results, and the mIoU value could reach more than 0.97. It is hoped that the method could be conducive to the automatic auxiliary diagnosis of blood diseases in the future.
High resolution (HR) magnetic resonance images (MRI) or computed tomography (CT) images can provide clearer anatomical details of human body, which facilitates early diagnosis of the diseases. However, due to the imaging system, imaging environment and human factors, it is difficult to obtain clear high-resolution images. In this paper, we proposed a novel medical image super resolution (SR) reconstruction method via multi-scale information distillation (MSID) network in the non-subsampled shearlet transform (NSST) domain, namely NSST-MSID network. We first proposed a MSID network that mainly consisted of a series of stacked MSID blocks to fully exploit features from images and effectively restore the low resolution (LR) images to HR images. In addition, most previous methods predict the HR images in the spatial domain, producing over-smoothed outputs while losing texture details. Thus, we viewed the medical image SR task as the prediction of NSST coefficients, which make further MSID network keep richer structure details than that in spatial domain. Finally, the experimental results on our constructed medical image datasets demonstrated that the proposed method was capable of obtaining better peak signal to noise ratio (PSNR), structural similarity (SSIM) and root mean square error (RMSE) values and keeping global topological structure and local texture detail better than other outstanding methods, which achieves good medical image reconstruction effect.
Most current medical image segmentation models are primarily built upon the U-shaped network (U-Net) architecture, which has certain limitations in capturing both global contextual information and fine-grained details. To address this issue, this paper proposes a novel U-shaped network model, termed the Multi-View U-Net (MUNet), which integrates self-attention and multi-view attention mechanisms. Specifically, a newly designed multi-view attention module is introduced to aggregate semantic features from different perspectives, thereby enhancing the representation of fine details in images. Additionally, the MUNet model leverages a self-attention encoding block to extract global image features, and by fusing global and local features, it improves segmentation performance. Experimental results demonstrate that the proposed model achieves superior segmentation performance in coronary artery image segmentation tasks, significantly outperforming existing models. By incorporating self-attention and multi-view attention mechanisms, this study provides a novel and efficient modeling approach for medical image segmentation, contributing to the advancement of intelligent medical image analysis.
The finite element method is a new method to study the mechanism of brain injury caused by blunt instruments. But it is not easy to be applied because of its technology barrier of time-consuming and strong professionalism. In this study, a rapid and quantitative evaluation method was investigated to analyze the craniocerebral injury induced by blunt sticks based on convolutional neural network and finite element method. The velocity curve of stick struck and the maximum principal strain of brain tissue (cerebrum, corpus callosum, cerebellum and brainstem) from the finite element simulation were used as the input and output parameters of the convolutional neural network The convolutional neural network was trained and optimized by using the 10-fold cross-validation method. The Mean Absolute Error (MAE), Mean Square Error (MSE), and Goodness of Fit (R2) of the finally selected convolutional neural network model for the prediction of the maximum principal strain of the cerebrum were 0.084, 0.014, and 0.92, respectively. The predicted results of the maximum principal strain of the corpus callosum were 0.062, 0.007, 0.90, respectively. The predicted results of the maximum principal strain of the cerebellum and brainstem were 0.075, 0.011, and 0.94, respectively. These results show that the research and development of the deep convolutional neural network can quickly and accurately assess the local brain injury caused by the sticks blow, and have important application value for understanding the quantitative evaluation and the brain injury caused by the sticks struck. At the same time, this technology improves the computational efficiency and can provide a basis reference for transforming the current acceleration-based brain injury research into a focus on local brain injury research.
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.
Deep learning method can be used to automatically analyze electrocardiogram (ECG) data and rapidly implement arrhythmia classification, which provides significant clinical value for the early screening of arrhythmias. How to select arrhythmia features effectively under limited abnormal sample supervision is an urgent issue to address. This paper proposed an arrhythmia classification algorithm based on an adaptive multi-feature fusion network. The algorithm extracted RR interval features from ECG signals, employed one-dimensional convolutional neural network (1D-CNN) to extract time-domain deep features, employed Mel frequency cepstral coefficients (MFCC) and two-dimensional convolutional neural network (2D-CNN) to extract frequency-domain deep features. The features were fused using adaptive weighting strategy for arrhythmia classification. The paper used the arrhythmia database jointly developed by the Massachusetts Institute of Technology and Beth Israel Hospital (MIT-BIH) and evaluated the algorithm under the inter-patient paradigm. Experimental results demonstrated that the proposed algorithm achieved an average precision of 75.2%, an average recall of 70.1% and an average F1-score of 71.3%, demonstrating high classification accuracy and being able to provide algorithmic support for arrhythmia classification in wearable devices.
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.
This study aims to address the limitations in gesture recognition caused by the susceptibility of temporal and frequency domain feature extraction from surface electromyography signals, as well as the low recognition rates of conventional classifiers. A novel gesture recognition approach was proposed, which transformed surface electromyography signals into grayscale images and employed convolutional neural networks as classifiers. The method began by segmenting the active portions of the surface electromyography signals using an energy threshold approach. Temporal voltage values were then processed through linear scaling and power transformations to generate grayscale images for convolutional neural network input. Subsequently, a multi-view convolutional neural network model was constructed, utilizing asymmetric convolutional kernels of sizes 1 × n and 3 × n within the same layer to enhance the representation capability of surface electromyography signals. Experimental results showed that the proposed method achieved recognition accuracies of 98.11% for 13 gestures and 98.75% for 12 multi-finger movements, significantly outperforming existing machine learning approaches. The proposed gesture recognition method, based on surface electromyography grayscale images and multi-view convolutional neural networks, demonstrates simplicity and efficiency, substantially improving recognition accuracy and exhibiting strong potential for practical applications.