Short-term prognostic predictive value of deep-learning assisted quantitative myocardial contrast echocardiography in ST-elevated myocardial infarction after primary percutaneous coronary intervention / 中华超声影像学杂志

Objective:To explore the prognostic predictive value of deep neural network (DNN) assisted myocardial contrast echocardiography (MCE) quantitative analysis of ST-elevated myocardial infarction (STEMI) patients after successful percutaneous coronary intervention(PCI).Methods:A retrospective analysis was performed in 97 STEMI patients with thrombolysis in myocardial infarction-3 flow in infarct vessel after primary PCI in Renmin Hospital of Wuhan University from June to November 2021. MCE was performed within 48 h after PCI. Patients were followed up to 120 days. The adverse events were defined as cardiac death, hospitalization for congestive heart failure, reinfarction, stroke and recurrent angina. The framework consisted of the U-net and hierarchical convolutional LSTMs. The plateau myocardial contrast intensity (A), micro-bubble rate constant (β), and microvascular blood flow (MBF) for all myocardial segments were obtained by the framework, and then underwent variability analysis. Patients were divided into low MBF group and high MBF group based on MBF values, the baseline characteristics and adverse events were compared between the two groups. Other variables included biomarkers, ventricular wall motion analysis, MCE qualitative analysis, and left ventricular ejection fraction. The relationship between various variables and prognosis was investigated using Cox regression analysis. The ROC curve was plotted to evaluate the diagnostic efficacy of the models, and the diagnostic efficacy of the models was compared using the integrated discrimination improvement index (IDI).Results:The time-cost for processing all 3 810 frames from 97 patients was 377 s. 92.89% and 7.11% of the frames were evaluated by an experienced echocardiographer as "good segmentation" and "correction needed". The correlation coefficients of A, β, and MBF ranged from 0.97 to 0.99 for intra-observer and inter-observer variability. During follow-up, 20 patients met the adverse events. Multivariate Cox regression analysis showed that for each increase of 1 IU/s in MBF of the infarct-related artery territory, the risk of adverse events decreased by 6% ( HR 0.94, 95% CI =0.91-0.98). There was a 4.5-fold increased risk of adverse events in the low MBF group ( HR 5.50, 95% CI=1.55-19.49). After incorporating DNN-assisted MCE quantitative analysis into qualitative analysis, the IDI for prognostic prediction was 15% (AUC 0.86, sensitivity 0.78, specificity 0.73). Conclusions:MBF of the area supplied by infarct-related artery after STEMI-PCI is an independent protective factor for short-term prognosis. The DNN-assisted MCE quantitative analysis is an objective, efficient, and reproducible method to evaluate microvascular perfusion. Assessment of culprit-MBF after PCI in STEMI patients adds independent short-term prognostic information over qualitative analysis.It has the potential to be a valuable tool for risk stratification and clinical follow-up.

Comparison of prediction ability of two extended Cox models in nonlinear survival data analysis / 南方医科大学学报

OBJECTIVE@#To compare the predictive ability of two extended Cox models in nonlinear survival data analysis.@*METHODS@#Through Monte Carlo simulation and empirical study and with the conventional Cox Proportional Hazards model and Random Survival Forests as the reference models, we compared restricted cubic spline Cox model (Cox_RCS) and DeepSurv neural network Cox model (Cox_DNN) for their prediction ability in nonlinear survival data analysis. Concordance index was used to evaluate the differentiation of the prediction results (a larger concordance index indicates a better prediction ability of the model). Integrated Brier Score was used to evaluate the calibration degree of the prediction (a smaller index indicates a better prediction ability).@*RESULTS@#For data that met requirement of the proportion risk, the Cox_RCS model had the best prediction ability regardless of the sample size or deletion rate. For data that failed to meet the proportion risk, the prediction ability of Cox_DNN was optimal for a large sample size (≥500) with a low deletion (< 40%); the prediction ability of Cox_RCS was superior to those of other models in all other scenarios. For example data, the Cox_RCS model showed the best performance.@*CONCLUSION@#In analysis of nonlinear low maintenance data, Cox_RCS and Cox_DNN have their respective advantages and disadvantages in prediction. The conventional survival analysis methods are not inferior to machine learning or deep learning methods under certain conditions.

Hearing recovery prediction and prognostic factors of idiopathic sudden sensorineural hearing loss: a retrospective analysis with a deep neural network model

Abstract Objective Idiopathic Sudden Sensorineural Hearing Loss (ISSHL) is an otologic emergency, and an early prediction of prognosis may facilitate proper treatment. Therefore, we investigated the prognostic factors for predicting the recovery in patients with ISSHL treated with combined treatment method using machine learning models. Methods We retrospectively reviewed the medical records of 298 patients with ISSHL at a tertiary medical institution between January 2015 and September 2020. Fifty-two variables were analyzed to predict hearing recovery. Recovery was defined using Siegel's criteria, and the patients were categorized into recovery and non-recovery groups. Recovery was predicted by various machine learning models. In addition, the prognostic factors were analyzed using the difference in the loss function. Results There were significant differences in variables including age, hypertension, previous hearing loss, ear fullness, duration of hospital admission, initial hearing level of the affected and unaffected ears, and post-treatment hearing level between recovery and non-recovery groups. The deep neural network model showed the highest predictive performance (accuracy, 88.81%; area under the receiver operating characteristic curve, 0.9448). In addition, initial hearing level of affected and non-affected ear, post-treatment (2-weeks) hearing level of affected ear were significant factors for predicting the prognosis. Conclusion The deep neural network model showed the highest predictive performance for recovery in patients with ISSHL. Some factors with prognostic value were identified. Further studies using a larger patient population are warranted. Level of evidence: Level 4.

Hybrid deep learning model for risk prediction of fracture in patients with diabetes and osteoporosis / 医学前沿

The fracture risk of patients with diabetes is higher than those of patients without diabetes due to hyperglycemia, usage of diabetes drugs, changes in insulin levels, and excretion, and this risk begins as early as adolescence. Many factors including demographic data (such as age, height, weight, and gender), medical history (such as smoking, drinking, and menopause), and examination (such as bone mineral density, blood routine, and urine routine) may be related to bone metabolism in patients with diabetes. However, most of the existing methods are qualitative assessments and do not consider the interactions of the physiological factors of humans. In addition, the fracture risk of patients with diabetes and osteoporosis has not been further studied previously. In this paper, a hybrid model combining XGBoost with deep neural network is used to predict the fracture risk of patients with diabetes and osteoporosis, and investigate the effect of patients' physiological factors on fracture risk. A total of 147 raw input features are considered in our model. The presented model is compared with several benchmarks based on various metrics to prove its effectiveness. Moreover, the top 18 influencing factors of fracture risks of patients with diabetes are determined.

Optimization of polysaccharide extraction from Hippocampus by deep neural network and Box-Behnken design-response surface methodology / 中国中药杂志

In this paper, the extraction rate of crude polysaccharides and the yield of polysaccharides from Hippocampus served as test indicators. The comprehensive evaluation indicators were assigned by the R language combined with the entropy weight method. The Box-Behnken design-response surface methodology(BBD-RSM) and the deep neural network(DNN) were employed to screen the optimal parameters for the polysaccharide extraction from Hippocampus. These two modeling methods were compared and verified experimentally for the process optimization. This study provides a reference for the industrialization of effective component extraction from Chinese medicinals and achieves the effective combination of modern technology and traditional Chinese medicine.

Prediction of drug-induced cell viability by SAE-XGBoost algorithm based on LINCS-L1000 perturbation signal / 生物工程学报

Different cell lines have different perturbation signals in response to specific compounds, and it is important to predict cell viability based on these perturbation signals and to uncover the drug sensitivity hidden underneath the phenotype. We developed an SAE-XGBoost cell viability prediction algorithm based on the LINCS-L1000 perturbation signal. By matching and screening three major dataset, LINCS-L1000, CTRP and Achilles, a stacked autoencoder deep neural network was used to extract the gene information. These information were combined with the RW-XGBoost algorithm to predict the cell viability under drug induction, and then to complete drug sensitivity inference on the NCI60 and CCLE datasets. The model achieved good results compared to other methods with a Pearson correlation coefficient of 0.85. It was further validated on an independent dataset, corresponding to a Pearson correlation coefficient of 0.68. The results indicate that the proposed method can help discover novel and effective anti-cancer drugs for precision medicine.

Comparison of Machine Learning Methods Applied to Estimation of Side Effect in Drug Interaction Using Japanese Adverse Drug Event Report (JADER) Database / 医薬品情報学

Objective: In this study, we analyzed the Japanese Adverse Drug Event Report (JADER) database in order detect unexpected adverse events using three polypharmacy machine learning models.Methods: The patient’s age, weight, height, gender, date and time of onset, subsequent appearance, and the taking medicines were preprocessed. They were applied for the prediction of adverse events using three machine learning procedures such as support vector machine (SVM), deep neural network (DNN) and random forest (RF).Results: Precision, matching, reproduction and F-values were almost same between the three techniques. Polypharmacy effects were predicted in approximately 80% of adverse events. Unexpected predictions were observed between DNN and RF, but different from SVM.Conclusion: Results suggest that the combination of DNN or RF and SVM can yield accurate predictions. We also suggest that RF is more useful because of its easy validation.

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.

Preoperative CT angiography and intraoperative X-ray image registration algorithm for thoracic aortic endovascular repair / 中国组织工程研究

BACKGROUND: Thoracic aortic endovascular repair is an important method for treating aortic dissection and thoracic aortic aneurysm. The success of the operation depends on whether the stent graft is placed in the correct position. However, when the stent is implanted, the aorta in the intraoperative X-ray image is invisible, so the operation is difficult and the risk is high. Registration of preoperative CT angiography and intraoperative X-ray images can help doctors place stents and increase success rates. OBJECTIVE: To propose a preoperative CT angiography and intraoperative X-ray image registration algorithm for thoracic aortic endovascular repair. METHODS: Firstly, digital reconstruction images of CT angiography and bone CT were performed under different virtual perspectives, and the two were superimposed to obtain a digital reconstruction image library under various angles of position and orientation for intraoperative X-ray images. Secondly, we proposed a deep neural network based on branch decoding structure. Using digital reconstruction image library training, the position and attitude parameters of intraoperative X-ray images could be estimated to obtain CT angiography and intraoperative X-ray images. The spatial positional relationship was obtained. Finally, according to the pose parameters of the X-ray image in the CT angiography coordinate system, the thoracic aorta image in the CT angiography was re-projected and superimposed into the intraoperative X-ray image to navigation assistance for the doctors. RESULTSANDCONCLUSION: (1) The experimental results show that the root mean square error of the proposed algorithmis reduced by 17%comparedwiththe traditional algorithmsof gradient correlation and mode strength. (2) In the dual-branch code structure network, the parameter estimation error is reduced to 30% of the network without branching structure in the digital reconstruction image test set. (3) In the experimental X-image experiment, the root mean square error is also reduced by2%.

Discussion of artificial intelligence application in medical imaging / 第二军医大学学报

As a new generation of artificial intelligence technology, the deep neural network takes the cognitive ability of machine to a historical high level in natural language processing, learning ability and computer vision. At present, the application of deep neural network in medical imaging can be categorized into discovery of anomalies, quantitative measurement, and differential diagnosis. Medical imaging research based on deep neural network research has involved various medical imaging domains such as radiological imaging, pathological images, ultrasound imaging, and endoscopic imaging. In several tasks, deep neural network has demonstrated physician-level or even above-physician-level performance. In the context of rapid development of artificial intelligence in imaging medicine, physicians should adopt a more objective, scientific, and proactive attitude towards artificial intelligence technology, and become the masters of artificial intelligence technology and the creators of a futuristic medical world assisted by artificial intelligence technology.

Deep-learning-based automatic computer-aided diagnosis system for diabetic retinopathy

The high-pace rise in advanced computing and imaging systems has given rise to a new research dimension called computer-aided diagnosis (CAD) system for various biomedical purposes. CAD-based diabetic retinopathy (DR) can be of paramount significance to enable early disease detection and diagnosis decision. Considering the robustness of deep neural networks (DNNs) to solve highly intricate classification problems, in this paper, AlexNet DNN, which functions on the basis of convolutional neural network (CNN), has been applied to enable an optimal DR CAD solution. The DR model applies a multilevel optimization measure that incorporates pre-processing, adaptive-learning-based Gaussian mixture model (GMM)-based concept region segmentation, connected component-analysis-based region of interest (ROI) localization, AlexNet DNN-based highly dimensional feature extraction, principle component analysis (PCA)- and linear discriminant analysis (LDA)-based feature selection, and support-vector-machine-based classification to ensure optimal five-class DR classification. The simulation results with standard KAGGLE fundus datasets reveal that the proposed AlexNet DNN-based DR exhibits a better performance with LDA feature selection, where it exhibits a DR classification accuracy of 97.93% with FC7 features, whereas with PCA, it shows 95.26% accuracy. Comparative analysis with spatial invariant feature transform (SIFT) technique (accuracy—94.40%) based DR feature extraction also confirms that AlexNet DNN-based DR outperforms SIFT-based DR.

The research of sleep staging based on single-lead electrocardiogram and deep neural network

The polysomnogram (PSG) analysis is considered the golden standard for sleep staging under the clinical environment. The electroencephalogram (EEG) signal is the most important signal for classification of sleep stages. However, in-vivo signal recording and analysis of EEG signal presents us with a few technical challenges. Electrocardiogram signals on the other hand, are easier to record, and can provide an attractive alternative for home sleep monitoring. In this paper we describe a method based on deep neural network (DNN), which can be used for the classification of the sleep stages into Wake (W), rapid-eye-movement (REM) and non-rapid-eye-movement (NREM) sleep stage. We apply the sleep stage stacked autoencoder to constitute a 4-layer DNN model. In order to test the accuracy of our method, eighteen PSGs from the MIT-BIH Polysomnographic Database were used. A total of 11 features were extracted from each electrocardiogram recording The experimental design employs cross-validation across subjects, ensuring the independence of the training and the test data. We obtained an accuracy of 77% and a Cohen's kappa coefficient of about 0.56 for the classification of Wake, REM and NREM.