An automatic diagnostic system for the urodynamic study applying in lower urinary tract dysfunction.

OBJECTIVE: To establish an automatic diagnostic system based on machine learning for preliminarily analysis of urodynamic study applying in lower urinary tract dysfunction (LUTD). METHODS: The eight most common conditions of LUTDs were included in the present study. A total of 527 eligible patients with complete data, from the year of 2015 to 2020, were enrolled in this study. In total, two global parameters (patients' age and sex) and 13 urodynamic parameters were considered to be the input for machine learning algorithms. Three machine learning approaches were applied and evaluated in this study, including Decision Tree (DT), Logistic Regression (LR), and Support Vector Machine (SVM). RESULTS: By applying machine learning algorithms into the 8 common LUTDs, the DT models achieved the AUC of 0.63-0.98, the LR models achieved the AUC of 0.73-0.99, and the SVM models achieved the AUC of 0.64-1.00. For mutually exclusive diagnoses of underactive detrusor and acontractile detrusor, we developed a classification model that classifies the patients into either of these two diseases or double-negative class. For this classification method, the DT models achieved the AUC of 0.82-0.85 and the SVM models achieved the AUC of 0.86-0.90. Among all these models, the LR and the SVM models showed better performance. The best model of these diagnostic tasks achieved an average AUC of 0.90 (0.90 ± 0.08). CONCLUSIONS: An automatic diagnostic system was developed using three machine learning models in urodynamic studies. This automated machine learning process could lead to promising assistance and enhancements of diagnosis and provide more useful reference for LUTD treatment.

DAIS: Automatic Channel Pruning via Differentiable Annealing Indicator Search.

The convolutional neural network (CNN) has achieved great success in fulfilling computer vision tasks despite large computation overhead against efficient deployment. Channel pruning is usually applied to reduce the model redundancy while preserving the network structure, such that the pruned network can be easily deployed in practice. However, existing channel pruning methods require hand-crafted rules, which can result in a degraded model performance with respect to the tremendous potential pruning space given large neural networks. In this article, we introduce differentiable annealing indicator search (DAIS) that leverages the strength of neural architecture search in the channel pruning and automatically searches for the effective pruned model with given constraints on computation overhead. Specifically, DAIS relaxes the binarized channel indicators to be continuous and then jointly learns both indicators and model parameters via bi-level optimization. To bridge the non-negligible discrepancy between the continuous model and the target binarized model, DAIS proposes an annealing-based procedure to steer the indicator convergence toward binarized states. Moreover, DAIS designs various regularizations based on a priori structural knowledge to control the pruning sparsity and to improve model performance. Experimental results show that DAIS outperforms state-of-the-art pruning methods on CIFAR-10, CIFAR-100, and ImageNet.

Scalp-recorded high-frequency oscillations in childhood epileptic encephalopathy with continuous spike-and-wave during sleep with different etiologies.

OBJECTIVE: To investigate high-frequency oscillations (HFOs) in epileptic encephalopathy with continuous spike-and-wave during sleep (CSWS) with different etiologies. METHODS: Twenty-one CSWS patients treated with methylprednisolone were divided into structural group and genetic/unknown group. Comparisons were made between the two etiological groups: selected clinical variables including gender, age parameters, seizure frequencies and antiepileptic drugs; distribution of HFOs in pre-methylprednisolone electroencephalography (EEG) and percentage changes of HFOs and spikes after methylprednisolone treatment. RESULTS: There were 7 patients (33%) in structural group and 14 patients (68%) in genetic/unknown group. No significant difference was found between the two groups regarding selected clinical variables. HFOs were found in 12 patients in pre-methylprednisolone EEG. The distribution of HFOs was focal and accordant with lesions in 5 of structural group, and it was also focal but in different brain regions in 7 of genetic/unknown group. The percentage reduction of total HFOs and spikes was 81% (158/195) and 19% (1956/10,037) in structural group, while 98% (315/323) and 55% (6658/12,258) in genetic/unknown group after methylprednisolone treatment. CONCLUSION: The etiologies had no distinct correlation with some clinical characteristics in CSWS. HFOs recorded on scalp EEG might not only be used as makers of seizure-onset zone (SOZ), but also have association with functional disruption of brain networks. Both HFOs and spikes reduced more in genetic/unknown patients than that in structural patients after methylprednisolone treatment and HFOs were more sensitive to treatment than spikes.

WE-CARE: an intelligent mobile telecardiology system to enable mHealth applications.

Recently, cardiovascular disease (CVD) has become one of the leading death causes worldwide, and it contributes to 41% of all deaths each year in China. This disease incurs a cost of more than 400 billion US dollars in China on the healthcare expenditures and lost productivity during the past ten years. It has been shown that the CVD can be effectively prevented by an interdisciplinary approach that leverages the technology development in both IT and electrocardiogram (ECG) fields. In this paper, we present WE-CARE , an intelligent telecardiology system using mobile 7-lead ECG devices. Because of its improved mobility result from wearable and mobile ECG devices, the WE-CARE system has a wider variety of applications than existing resting ECG systems that reside in hospitals. Meanwhile, it meets the requirement of dynamic ECG systems for mobile users in terms of the detection accuracy and latency. We carried out clinical trials by deploying the WE-CARE systems at Peking University Hospital. The clinical results clearly showed that our solution achieves a high detection rate of over 95% against common types of anomalies in ECG, while it only incurs a small detection latency around one second, both of which meet the criteria of real-time medical diagnosis. As demonstrated by the clinical results, the WE-CARE system is a useful and efficient mHealth (mobile health) tool for the cardiovascular disease diagnosis and treatment in medical platforms.