Automated detection of sleep-arousal using multi-scale convolution and self-attention mechanism / 生物医学工程学杂志

In clinical, manually scoring by technician is the major method for sleep arousal detection. This method is time-consuming and subjective. This study aimed to achieve an end-to-end sleep-arousal events detection by constructing a convolutional neural network based on multi-scale convolutional layers and self-attention mechanism, and using 1 min single-channel electroencephalogram (EEG) signals as its input. Compared with the performance of the baseline model, the results of the proposed method showed that the mean area under the precision-recall curve and area under the receiver operating characteristic were both improved by 7%. Furthermore, we also compared the effects of single modality and multi-modality on the performance of the proposed model. The results revealed the power of single-channel EEG signals in automatic sleep arousal detection. However, the simple combination of multi-modality signals may be counterproductive to the improvement of model performance. Finally, we also explored the scalability of the proposed model and transferred the model into the automated sleep staging task in the same dataset. The average accuracy of 73% also suggested the power of the proposed method in task transferring. This study provides a potential solution for the development of portable sleep monitoring and paves a way for the automatic sleep data analysis using the transfer learning method.

Non-rigid registration for medical images based on deformable convolution and multi-scale feature focusing modules / 生物医学工程学杂志

Non-rigid registration plays an important role in medical image analysis. U-Net has been proven to be a hot research topic in medical image analysis and is widely used in medical image registration. However, existing registration models based on U-Net and its variants lack sufficient learning ability when dealing with complex deformations, and do not fully utilize multi-scale contextual information, resulting insufficient registration accuracy. To address this issue, a non-rigid registration algorithm for X-ray images based on deformable convolution and multi-scale feature focusing module was proposed. First, it used residual deformable convolution to replace the standard convolution of the original U-Net to enhance the expression ability of registration network for image geometric deformations. Then, stride convolution was used to replace the pooling operation of the downsampling operation to alleviate feature loss caused by continuous pooling. In addition, a multi-scale feature focusing module was introduced to the bridging layer in the encoding and decoding structure to improve the network model's ability of integrating global contextual information. Theoretical analysis and experimental results both showed that the proposed registration algorithm could focus on multi-scale contextual information, handle medical images with complex deformations, and improve the registration accuracy. It is suitable for non-rigid registration of chest X-ray images.

Prediction of pulp exposure risk of carious pulpitis based on deep learning / 华西口腔医学杂志

OBJECTIVES@#This study aims to predict the risk of deep caries exposure in radiographic images based on the convolutional neural network model, compare the prediction results of the network model with those of senior dentists, evaluate the performance of the model for teaching and training stomatological students and young dentists, and assist dentists to clarify treatment plans and conduct good doctor-patient communication before surgery.@*METHODS@#A total of 206 cases of pulpitis caused by deep caries were selected from the Department of Stomatological Hospital of Tianjin Medical University from 2019 to 2022. According to the inclusion and exclusion criteria, 104 cases of pulpitis were exposed during the decaying preparation period and 102 cases of pulpitis were not exposed. The 206 radiographic images collected were randomly divided into three groups according to the proportion: 126 radiographic images in the training set, 40 radiographic images in the validation set, and 40 radiographic images in the test set. Three convolutional neural networks, visual geometry group network (VGG), residual network (ResNet), and dense convolutional network (DenseNet) were selected to analyze the rules of the radiographic images in the training set. The radiographic images of the validation set were used to adjust the super parameters of the network. Finally, 40 radiographic images of the test set were used to evaluate the performance of the three network models. A senior dentist specializing in dental pulp was selected to predict whether the deep caries of 40 radiographic images in the test set were exposed. The gold standard is whether the pulp is exposed after decaying the prepared hole during the clinical operation. The prediction effect of the three network models (VGG, ResNet, and DenseNet) and the senior dentist on the pulp exposure of 40 radiographic images in the test set were compared using receiver operating characteristic (ROC) curve, area under the ROC curve (AUC), accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score to select the best network model.@*RESULTS@#The best network model was DenseNet model, with AUC of 0.97. The AUC values of the ResNet model, VGG model, and the senior dentist were 0.89, 0.78, and 0.87, respectively. Accuracy was not statistically different between the senior dentist (0.850) and the DenseNet model (0.850)(P>0.05). Kappa consistency test showed moderate reliability (Kappa=0.6>0.4, P<0.05).@*CONCLUSIONS@#Among the three convolutional neural network models, the DenseNet model has the best predictive effect on whether deep caries are exposed in imaging. The predictive effect of this model is equivalent to the level of senior dentists specializing in dental pulp.

Novel Pulmonary Nodule Position Detection Method Based on Multiscale Convolution / 中国医疗器械杂志

OBJECTIVE@#In order to improve the accuracy of the current pulmonary nodule location detection method based on CT images, reduce the problem of missed detection or false detection, and effectively assist imaging doctors in the diagnosis of pulmonary nodules.@*METHODS@#Propose a novel method for detecting the location of pulmonary nodules based on multiscale convolution. First, image preprocessing methods are used to eliminate the noise and artifacts in lung CT images. Second, multiple adjacent single-frame CT images are selected to be concatenate into multi-frame images, and the feature extraction is carried out through the artificial neural network model U-Net improved by multi-scale convolution to enhanced feature extraction capability for pulmonary nodules of different sizes and shapes, so as to improve the accuracy of feature extraction of pulmonary nodules. Finally, using point detection to improve the loss function of U-Net training process, the accuracy of pulmonary nodule location detection is improved.@*RESULTS@#The accuracy of detecting pulmonary nodules equal or larger than 3 mm and smaller than 3 mm are 98.02% and 96.94% respectively.@*CONCLUSIONS@#This method can effectively improve the detection accuracy of pulmonary nodules on CT image sequence, and can better meet the diagnostic needs of pulmonary nodules.

Image segmentation of skin lesions based on dense atrous spatial pyramid pooling and attention mechanism / 生物医学工程学杂志

The skin is the largest organ of the human body, and many visceral diseases will be directly reflected on the skin, so it is of great clinical significance to accurately segment the skin lesion images. To address the characteristics of complex color, blurred boundaries, and uneven scale information, a skin lesion image segmentation method based on dense atrous spatial pyramid pooling (DenseASPP) and attention mechanism is proposed. The method is based on the U-shaped network (U-Net). Firstly, a new encoder is redesigned to replace the ordinary convolutional stacking with a large number of residual connections, which can effectively retain key features even after expanding the network depth. Secondly, channel attention is fused with spatial attention, and residual connections are added so that the network can adaptively learn channel and spatial features of images. Finally, the DenseASPP module is introduced and redesigned to expand the perceptual field size and obtain multi-scale feature information. The algorithm proposed in this paper has obtained satisfactory results in the official public dataset of the International Skin Imaging Collaboration (ISIC 2016). The mean Intersection over Union (mIOU), sensitivity (SE), precision (PC), accuracy (ACC), and Dice coefficient (Dice) are 0.901 8, 0.945 9, 0.948 7, 0.968 1, 0.947 3, respectively. The experimental results demonstrate that the method in this paper can improve the segmentation effect of skin lesion images, and is expected to provide an auxiliary diagnosis for professional dermatologists.

Segmentation of ground glass pulmonary nodules using full convolution residual network based on atrous spatial pyramid pooling structure and attention mechanism / 生物医学工程学杂志

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.

An EHG-based Preterm Delivery Prediction Algorithm via Convolution Neural Network / 中国医疗器械杂志

Premature delivery is one of the direct factors that affect the early development and safety of infants. Its direct clinical manifestation is the change of uterine contraction intensity and frequency. Uterine Electrohysterography(EHG) signal collected from the abdomen of pregnant women can accurately and effectively reflect the uterine contraction, which has higher clinical application value than invasive monitoring technology such as intrauterine pressure catheter. Therefore, the research of fetal preterm birth recognition algorithm based on EHG is particularly important for perinatal fetal monitoring. We proposed a convolution neural network(CNN) based on EHG fetal preterm birth recognition algorithm, and a deep CNN model was constructed by combining the Gramian angular difference field(GADF) with the transfer learning technology. The structure of the model was optimized using the clinical measured term-preterm EHG database. The classification accuracy of 94.38% and F1 value of 97.11% were achieved. The experimental results showed that the model constructed in this paper has a certain auxiliary diagnostic value for clinical prediction of premature delivery.

The Effects of Different Adaptive Statistical Iterative Reconstruction-V and Convolution Kernel Parameters on Auto-Segmentation Stability in CT Images / 中国医疗器械杂志

Objective The study aims to investigate the effects of different adaptive statistical iterative reconstruction-V（ ASiR-V） and convolution kernel parameters on stability of CT auto-segmentation which is based on deep learning. Method Twenty patients who have received pelvic radiotherapy were selected and different reconstruction parameters were used to establish CT images dataset. Then structures including three soft tissue organs (bladder, bowelbag, small intestine) and five bone organs (left and right femoral head, left and right femur, pelvic) were segmented automatically by deep learning neural network. Performance was evaluated by dice similarity coefficient（ DSC） and Hausdorff distance, using filter back projection(FBP) as the reference. Results Auto-segmentation of deep learning is greatly affected by ASIR-V, but less affected by convolution kernel, especially in soft tissues. Conclusion The stability of auto-segmentation is affected by parameter selection of reconstruction algorithm. In practical application, it is necessary to find a balance between image quality and segmentation quality, or improve segmentation network to enhance the stability of auto-segmentation.

A protein complex recognition method based on spatial-temporal graph convolution neural network / 南方医科大学学报

OBJECTIVE@#To propose a new method for mining complexes in dynamic protein network using spatiotemporal convolution neural network.@*METHODS@#The edge strength, node strength and edge existence probability are defined for modeling of the dynamic protein network. Based on the time series information and structure information on the graph, two convolution operators were designed using Hilbert-Huang transform, attention mechanism and residual connection technology to represent and learn the characteristics of the proteins in the network, and the dynamic protein network characteristic map was constructed. Finally, spectral clustering was used to identify the protein complexes.@*RESULTS@#The simulation results on several public biological datasets showed that the F value of the proposed algorithm exceeded 90% on DIP dataset and MIPS dataset. Compared with 4 other recognition algorithms (DPCMNE, GE-CFI, VGAE and NOCD), the proposed algorithm improved the recognition efficiency by 34.5%, 28.7%, 25.4% and 17.6%, respectively.@*CONCLUSION@#The application of deep learning technology can improve the efficiency in analysis of dynamic protein networks.

Application of a multi-task learning-based light-weight convolution neural network for the automatic segmentation of organs at risk in thorax / 中华放射肿瘤学杂志

Objective:To evaluate the application of a multi-task learning-based light-weight convolution neural network (MTLW-CNN) for the automatic segmentation of organs at risk (OARs) in thorax.Methods:MTLW-CNN consisted of several layers for sharing features and 3 branches for segmenting 3 OARs. 497 cases with thoracic tumors were collected. Among them, the computed tomography (CT) images encompassing lung, heart and spinal cord were included in this study. The corresponding contours delineated by experienced radiation oncologists were ground truth. All cases were randomly categorized into the training and validation set ( n=300) and test set ( n=197). By applying MTLW-CNN on the test set, the Dice similarity coefficients (DSCs) of 3 OARs, training and testing time and space complexity (S) were calculated and compared with those of Unet and DeepLabv3+ . To evaluate the effect of multi-task learning on the generalization performance of the model, 3 single-task light-weight CNNs (STLW-CNNs) were built. Their structures were totally the same as the corresponding branches in MTLW-CNN. After using the same data and algorithm to train STLW-CNN, the DSCs were statistically compared with MTLW-CNN on the testing set. Results:For MTLW-CNN, the averages (μ) of lung, heart and spinal cord DSCs were 0.954, 0.921 and 0.904, respectively. The differences of μ between MTLW-CNN and other two models (Unet and DeepLabv3+ ) were less than 0.020. The training and testing time of MTLW-CNN were 1/3 to 1/30 of that of Unet and DeepLabv3+ . S of MTLW-CNN was 1/42 of that of Unet and 1/1 220 of that of DeepLabv3+ . The differences of μ and standard deviation (σ) of lung and heart between MTLW-CNN and STLW-CNN were approximately 0.005 and 0.002. The difference of μ of spinal cord was 0.001, but σof STLW-CNN was 0.014 higher than that of MTLW-CNN.Conclusions:MTLW-CNN spends less time and space on high-precision automatic segmentation of thoracic OARs. It can improve the application efficiency and generalization performance of the models.

Automatic segmentation of kidney tumor based on cascaded multiscale convolutional neural networks / 生物医学工程学杂志

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.

Heart sound classification based on sub-band envelope and convolution neural network / 生物医学工程学杂志

Automatic classification of heart sounds plays an important role in the early diagnosis of congenital heart disease. A kind of heart sound classification algorithms based on sub-band envelope feature and convolution neural network was proposed in this paper, which did not need to segment the heart sounds according to cardiac cycle accurately. Firstly, the heart sound signal was divided into some frames. Then, the frame level heart sound signal was filtered with Gammatone filter bank to obtain the sub-band signals. Next, the sub-band envelope was extracted by Hilbert transform. After that, the sub-band envelope was stacked into a feature map. Finally, type Ⅰ and type Ⅱ convolution neural network were selected as classifier. The result shown that the sub-band envelope feature was better in type Ⅰ than type Ⅱ. The algorithm is tested with 1 000 heart sound samples. The test results show that the overall performance of the algorithm proposed in this paper is significantly improved compared with other similar algorithms, which provides a new method for automatic classification of congenital heart disease, and speeds up the process of automatic classification of heart sounds applied to the actual screening.

A preliminary study of a deep learning assisted diagnostic system with an artificial intelligence for detection of retina disease / 国际眼科杂志(Guoji Yanke Zazhi)

@#AIM:To evaluate the application value of artificial intelligence diagnosis system for fundus disease screening based on deep learning.<p>METHODS:A total of 1 345 patients(2 690 eyes)in our hospital were recruited from July 2018 to December 2018. The accuracy, specificity, consistency and sensitivity of the artificial intelligence diagnosis system were determined by comparison with ophthalmologist diagnosis and artificial intelligence diagnosis system which based on multi-layer deep convolution neural network learning. <p>RESULTS:The accuracy of artificial intelligence diagnosis system is 62.82%. There are 1-5(1.38±0.67)diagnoses among the patients, among which the accuracy of one diagnosis is 56.09%, the accuracy of two diagnosis is 77.96%, the accuracy of three diagnosis is 84.61%, the accuracy of four diagnosis is 86.95%, and the accuracy of five diagnosis is 60.00%; The consistency kappa value without obvious abnormality and leopard pattern fundus was 0.044 and 0.169 respectively. The sensitivity was 3.00% and 99.6% respectively, the specificity was 99.7% and 14.2% respectively. The consistency Kappa value of other diagnosis was as high as 0.57-1.00, the sensitivity was as high as 65.1%-100%, and the specificity was as high as 93.0%-100%. <p>CONCLUSION:This study shows that the artificial intelligence diagnosis system based on multi-layer deep convolution neural network learning is a reliable alternative to diagnose retina diseases, and it is expected to become an effective screening tool for primary medical treatment.

Research on electrocardiogram classification using deep residual network with pyramid convolution structure / 生物医学工程学杂志

Recently, deep neural networks (DNNs) have been widely used in the field of electrocardiogram (ECG) signal classification, but the previous models have limited ability to extract features from raw ECG data. In this paper, a deep residual network model based on pyramidal convolutional layers (PC-DRN) was proposed to implement ECG signal classification. The pyramidal convolutional (PC) layer could simultaneously extract multi-scale features from the original ECG data. And then, a deep residual network was designed to train the classification model for arrhythmia detection. The public dataset provided by the physionet computing in cardiology challenge 2017(CinC2017) was used to validate the classification experiment of 4 types of ECG data. In this paper, the harmonic mean of classification accuracy and recall was selected as the evaluation indexes. The experimental results showed that the average sequence level ( ) of PC-DRN was improved from 0.857 to 0.920, and the average set level ( ) was improved from 0.876 to 0.925. Therefore, the PC-DRN model proposed in this paper provided a promising way for the feature extraction and classification of ECG signals, and provided an effective tool for arrhythmia classification.

Development of a Deep Learning Algorithm for Classification of Neuroblastoma / 中国医疗器械杂志

In this paper, the classification and location of neuroblastoma in NMR images are realized by using Deep Neural Network(CNN) algorithm as the core technology. The module is integrated to realize the development of computer-aided diagnostic software. It is used to make up for the gap in the field of intelligent identification and accurate positioning of neuroblastoma in the current nuclear magnetic resonance detection technology, effectively reduce the work intensity of doctors reading films, and further promote the clinical application and technical development of nuclear magnetic resonance detection technology in the diagnosis of neuroblastoma.

A review on brain age prediction in brain ageing / 生物医学工程学杂志

The human brain deteriorates as we age, and the rate and the trajectories of these changes significantly vary among brain regions and among individuals. Because neuroimaging data are potentially important indicators of individual's brain health, they are commonly used in brain age prediction. In this review, we summarize brain age prediction model from neuroimaging-based studies in the last ten years. The studies are categorized based on their image modalities and feature types. The results indicate that the prediction frameworks based on neuroimaging holds promise toward individualized brain age prediction. Finally, we addressed the challenges in brain age prediction and suggested some future research directions.

Heartbeat-based end-to-end classification of arrhythmias / 南方医科大学学报

OBJECTIVE@#We propose a heartbeat-based end-to-end classification of arrhythmias to improve the classification performance for supraventricular ectopic beat (SVEB) and ventricular ectopic beat (VEB).@*METHODS@#The ECG signals were preprocessed by heartbeat segmentation and heartbeat alignment. An arrhythmia classifier was constructed based on convolutional neural network, and the proposed loss function was used to train the classifier.@*RESULTS@#The proposed algorithm was verified on MIT-BIH arrhythmia database. The AUC of the proposed loss function for SVEB and VEB reached 0.77 and 0.98, respectively. With the first 5 min segment as the local data, the diagnostic sensitivities for SVEB and VEB were 78.28% and 98.88%, respectively; when 0, 50, 100, and 150 samples were used as the local data, the diagnostic sensitivities for SVEB and VEB reached 82.25% and 93.23%, respectively.@*CONCLUSIONS@#The proposed method effectively reduces the negative impact of class-imbalance and improves the diagnostic sensitivities for SVEB and VEB, and thus provides a new solution for automatic arrhythmia classification.

Segmentation of retinal image vessels based on fully convolutional network with depthwise separable convolution and channel weighting / 生物医学工程学杂志

Diseases such as diabetes and hypertension can lead to change the shape of the retinal blood vessels. Segmentation of fundus images is a key step in the process of quantitative analysis of the disease, which is instructive in the analysis and diagnosis of clinical diseases. In this paper, a method for the segmentation of retinal image vessels based on fully convolutional network (FCN) with depthwise separable convolution and channel weighting is presented. Firstly, CLAHE and Gamma correction of the green channel of the fundus image are used to enhance the contrast. Then, in order to adapt to network training, the enhanced image is divided into patches to expand the data. Finally, the depthwise separable convolution instead of the standard convolution method is used to increase the network width. Meanwhile, the channel weighting module is introduced to explicitly model the relationship between the characteristic channels in order to improve the distinguishability of the features. The combination of them is applied to the FCN and the results of expert manual identification are used to supervise the experiment on the DRIVE database. The results show that the segmentation accuracy of the proposed method in DRIVE database reached 0.963 0 and AUC reached 0.983 1. The segmentation accuracy in STARE database reached 0.962 0 and AUC achieved 0.983 0. To some extent, the proposed method has better feature resolution and better segmentation performance.

Method on automatic location of inserts in electron density phantom / 中国医学影像技术

Objective: To investigate automatic location of inserts in the electron density phantom (CIRS 062) based on deep neural network (DCNN). Methods Firstly, four inserts in CIRS 062 were segmented with DCNN model, namely the inhaled lung, the exhaled lung, the solid trabecular bone and the solid dense bone. Then Moore-neighbor tracking algorithm was used to process the segmentation results to obtain the precise segmentation edges. Finally, the other four inserts were located based on the geometric features. Results The results of Dice similarity coefficient were all >0.85, the precision were all >0.81, and F1-measure were all >0.61 based on DCNN. Conclusion The method based on DCNN can realize the automatic positioning of the inserts.

Super-resolution construction of intravascular ultrasound images using generative adversarial networks / 南方医科大学学报

The low-resolution ultrasound images have poor visual effects. Herein we propose a method for generating clearer intravascular ultrasound images based on super-resolution reconstruction combined with generative adversarial networks. We used the generative adversarial networks to generate the images by a generator and to estimate the authenticity of the images by a discriminator. Specifically, the low-resolution image was passed through the sub-pixel convolution layer -feature channels to generate -feature maps in the same size, followed by realignment of the corresponding pixels in each feature map into × sub-blocks, which corresponded to the sub-block in a high-resolution image; after amplification, an image with a -time resolution was generated. The generative adversarial networks can obtain a clearer image through continuous optimization. We compared the method (SRGAN) with other methods including Bicubic, super-resolution convolutional network (SRCNN) and efficient sub-pixel convolutional network (ESPCN), and the proposed method resulted in obvious improvements in the peak signal-to-noise ratio (PSNR) by 2.369 dB and in structural similarity index by 1.79% to enhance the diagnostic visual effects of intravascular ultrasound images.