Dechun Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dechun Zhao has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
Xiaorong Hou
College of Medical Informatics Chongqing Medical University Chongqing China
6 shared publicationsLu Qin
School of Life Health Information Science and Engineering, Chongqing University of Posts and Telecommunications, Chongqing, P.R. China
4 shared publicationsYuchen Shen
The School of Life Health Information Science and Engineering, Chongqing University of Posts and Telecommunications, China
4 shared publicationsLi Wei
School of Life Health Information Science and Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, P. R. China
3 shared publicationsYi Wang
College of bioinformatics, Chongqing University of Posts and Telecommunications, Chongqing, China
3 shared publicationsZiqiong Wang
School of Life Health Information Science and Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, P. R. China
3 shared publications
Latest Publications
- Open Access
- 5 January 2026
- Published by Frontiers Media SA in Frontiers in Neuroscience
- Vol. 19, 1707636
A wearable deep brain stimulation system for behavioral studies in rodents
Introduction: Deep brain stimulation (DBS) has been demonstrated to improve motor function by modulating brain circuits. However, the precise mechanism of action remains unclear. Existing DBS devices often lack suitability for rodent models, which are essential for exploring these mechanisms. The objective of this study is to design a wearable DBS device that enables more effective rodent behavior research. Methods: We developed a stimulator specifically designed for rodent behavioral research. The device features several key advantages, including low weight (5 g excluding battery), compact size (30 mm × 38 mm), and extended battery life (≥300 hours). Additionally, it integrates low-power Bluetooth, allowing for real-time wireless adjustment of stimulation parameters, such as frequency (80–150 Hz), pulse width (40–340 μs), and amplitude (2.3–3 V). Results: The stimulating effect of this device has been verified in a model of mild Parkinson’s disease (mild Parkinsonism) in rats, indicating that the device is effective in regulating the motor functions of rodents with Parkinson’s disease-like conditions. Discussion: This wearable DBS device shows broad applicability in behavior-based studies and offers a valuable tool for investigating the mechanisms of various brain nuclei in rodent models. It holds significant potential for enhancing the exploration of DBS effects in clinical and experimental settings.
- 5 January 2026
- Published by Frontiers Media SA in Frontiers in Neuroscience
- Vol. 19, 1707636
Open AccessA wearable deep brain stimulation system for behavioral studies in rodents
Introduction: Deep brain stimulation (DBS) has been demonstrated to improve motor function by modulating brain circuits. However, the precise mechanism of action remains unclear. Existing DBS devices often lack suitability for rodent models, which are essential for exploring these mechanisms. The objective of this study is to design a wearable DBS device that enables more effective rodent behavior research. Methods: We developed a stimulator specifically designed for rodent behavioral research. The device features several key advantages, including low weight (5 g excluding battery), compact size (30 mm × 38 mm), and extended battery life (≥300 hours). Additionally, it integrates low-power Bluetooth, allowing for real-time wireless adjustment of stimulation parameters, such as frequency (80–150 Hz), pulse width (40–340 μs), and amplitude (2.3–3 V). Results: The stimulating effect of this device has been verified in a model of mild Parkinson’s disease (mild Parkinsonism) in rats, indicating that the device is effective in regulating the motor functions of rodents with Parkinson’s disease-like conditions. Discussion: This wearable DBS device shows broad applicability in behavior-based studies and offers a valuable tool for investigating the mechanisms of various brain nuclei in rodent models. It holds significant potential for enhancing the exploration of DBS effects in clinical and experimental settings.
- 3 November 2025
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (2) , 547-563
MLAR-SleepNet: a automatic sleep staging model based on residual and multi-level attention network
- 3 November 2025
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (2) , 547-563
MLAR-SleepNet: a automatic sleep staging model based on residual and multi-level attention network
- 25 April 2025
- Vol. 42 (2) , 326-333
[Medical text classification model integrating medical entity label semantics].
Automatic classification of medical questions is of great significance in improving the quality and efficiency of online medical services, and belongs to the task of intent recognition. Joint entity recognition and intent recognition perform better than single task models. Currently, most publicly available medical text intent recognition datasets lack entity annotation, and manual annotation of these entities requires a lot of time and manpower. To solve this problem, this paper proposes a medical text classification model, bidirectional encoder representation based on transformer-recurrent convolutional neural network-entity-label-semantics (BRELS), which integrates medical entity label semantics. This model firstly utilizes an adaptive fusion mechanism to absorb prior knowledge of medical entity labels, achieving local feature enhancement. Then in global feature extraction, a lightweight recurrent convolutional neural network (LRCNN) is used to suppress parameter growth while preserving the original semantics of the text. The ablation and comparison experiments are conducted on three public medical text intent recognition datasets to validate the performance of the model. The results show that F1 score reaches 87.34%, 81.71%, and 77.74% on each dataset, respectively. The results show that the BRELS model can effectively identify and understand medical terminology, thereby effectively identifying users' intentions, which can improve the quality and efficiency of online medical services.
- 25 April 2025
- Vol. 42 (2) , 326-333
[Medical text classification model integrating medical entity label semantics].
Automatic classification of medical questions is of great significance in improving the quality and efficiency of online medical services, and belongs to the task of intent recognition. Joint entity recognition and intent recognition perform better than single task models. Currently, most publicly available medical text intent recognition datasets lack entity annotation, and manual annotation of these entities requires a lot of time and manpower. To solve this problem, this paper proposes a medical text classification model, bidirectional encoder representation based on transformer-recurrent convolutional neural network-entity-label-semantics (BRELS), which integrates medical entity label semantics. This model firstly utilizes an adaptive fusion mechanism to absorb prior knowledge of medical entity labels, achieving local feature enhancement. Then in global feature extraction, a lightweight recurrent convolutional neural network (LRCNN) is used to suppress parameter growth while preserving the original semantics of the text. The ablation and comparison experiments are conducted on three public medical text intent recognition datasets to validate the performance of the model. The results show that F1 score reaches 87.34%, 81.71%, and 77.74% on each dataset, respectively. The results show that the BRELS model can effectively identify and understand medical terminology, thereby effectively identifying users' intentions, which can improve the quality and efficiency of online medical services.
- 25 February 2025
- Vol. 42 (1) , 24-31
[Audiovisual emotion recognition based on a multi-head cross attention mechanism].
In audiovisual emotion recognition, representational learning is a research direction receiving considerable attention, and the key lies in constructing effective affective representations with both consistency and variability. However, there are still many challenges to accurately realize affective representations. For this reason, in this paper we proposed a cross-modal audiovisual recognition model based on a multi-head cross-attention mechanism. The model achieved fused feature and modality alignment through a multi-head cross-attention architecture, and adopted a segmented training strategy to cope with the modality missing problem. In addition, a unimodal auxiliary loss task was designed and shared parameters were used in order to preserve the independent information of each modality. Ultimately, the model achieved macro and micro F1 scores of 84.5% and 88.2%, respectively, on the crowdsourced annotated multimodal emotion dataset of actor performances (CREMA-D). The model in this paper can effectively capture intra- and inter-modal feature representations of audio and video modalities, and successfully solves the unity problem of the unimodal and multimodal emotion recognition frameworks, which provides a brand-new solution to the audiovisual emotion recognition.
- 25 February 2025
- Vol. 42 (1) , 24-31
[Audiovisual emotion recognition based on a multi-head cross attention mechanism].
In audiovisual emotion recognition, representational learning is a research direction receiving considerable attention, and the key lies in constructing effective affective representations with both consistency and variability. However, there are still many challenges to accurately realize affective representations. For this reason, in this paper we proposed a cross-modal audiovisual recognition model based on a multi-head cross-attention mechanism. The model achieved fused feature and modality alignment through a multi-head cross-attention architecture, and adopted a segmented training strategy to cope with the modality missing problem. In addition, a unimodal auxiliary loss task was designed and shared parameters were used in order to preserve the independent information of each modality. Ultimately, the model achieved macro and micro F1 scores of 84.5% and 88.2%, respectively, on the crowdsourced annotated multimodal emotion dataset of actor performances (CREMA-D). The model in this paper can effectively capture intra- and inter-modal feature representations of audio and video modalities, and successfully solves the unity problem of the unimodal and multimodal emotion recognition frameworks, which provides a brand-new solution to the audiovisual emotion recognition.
- 8 November 2024
- Published by SAGE Publications in Technology and Health Care
- Vol. 32 (6) , 4097-4107
State identification of Parkinson’s disease based on transfer learning
BACKGROUND: The local field potential (LFP) signals are a vital signal for studying the mechanisms of deep brain stimulation (DBS) and constructing adaptive DBS containing information related to the motor symptoms of Parkinson’s disease (PD). OBJECTIVE: A Parkinson’s disease state identification algorithm based on the feature extraction strategy of transfer learning was proposed. METHODS: The algorithm uses continuous wavelet transform (CWT) to convert one-dimensional LFP signals into two-dimensional gray-scalogram images and color images respectively, and designs a Bayesian optimized random forest (RF) classifier to replace the three fully connected layers for the classification task in the VGG16 model, to realize automatic identification of the pathological state of PD patients. RESULTS: It was found that consistently superior performance of gray-scalogram images over color images. The proposed algorithm achieved an accuracy of 97.76%, precision of 99.01%, recall of 96.47%, and F1-score of 97.73%, outperforming feature extractors such as VGG19, InceptionV3, ResNet50, and the lightweight network MobileNet. CONCLUSIONS: This algorithm has high accuracy and can distinguish the disease states of PD patients without manual feature extraction, effectively assisting the working of doctors.
- 8 November 2024
- Published by SAGE Publications in Technology and Health Care
- Vol. 32 (6) , 4097-4107
State identification of Parkinson’s disease based on transfer learning
BACKGROUND: The local field potential (LFP) signals are a vital signal for studying the mechanisms of deep brain stimulation (DBS) and constructing adaptive DBS containing information related to the motor symptoms of Parkinson’s disease (PD). OBJECTIVE: A Parkinson’s disease state identification algorithm based on the feature extraction strategy of transfer learning was proposed. METHODS: The algorithm uses continuous wavelet transform (CWT) to convert one-dimensional LFP signals into two-dimensional gray-scalogram images and color images respectively, and designs a Bayesian optimized random forest (RF) classifier to replace the three fully connected layers for the classification task in the VGG16 model, to realize automatic identification of the pathological state of PD patients. RESULTS: It was found that consistently superior performance of gray-scalogram images over color images. The proposed algorithm achieved an accuracy of 97.76%, precision of 99.01%, recall of 96.47%, and F1-score of 97.73%, outperforming feature extractors such as VGG19, InceptionV3, ResNet50, and the lightweight network MobileNet. CONCLUSIONS: This algorithm has high accuracy and can distinguish the disease states of PD patients without manual feature extraction, effectively assisting the working of doctors.
- 25 February 2024
- Vol. 41 (1) , 17-25
基于样本增强的帕金森病识别算法研究
帕金森病患者早期存在声带损伤,其声纹特征与健康人存在明显差异,可以利用该差异识别帕金森病,但帕金森病患者声纹数据样本不足,因此本文提出双自注意力深度卷积生成对抗网络模型进行样本增强,生成高分辨率的语谱图,进而采用深度学习方法进行帕金森病识别。该模型通过增加网络深度并结合梯度惩罚、频谱归一化技术改进样本的纹理清晰度,并且构建一个基于迁移学习的纯粹的卷积神经网络家族(ConvNeXt)作为分类网络,以此提取声纹特征并进行分类,提升了帕金森病识别准确率。在帕金森病语音数据集上进行本文算法有效性验证实验,对比样本增强前,本文所提模型生成的样本清晰度以及弗雷谢起始距离(FID)均得到提高,并且本文网络模型能够获得98.8%的准确率。本文研究结果表明,基于双自注意力深度卷积生成对抗网络样本增强的帕金森病识别算法能够准确区分健康人和帕金森病患者,有助于解决帕金森病早期识别声纹数据样本不足的问题。综上,本文方法有效提高小样本帕金森病语音数据集分类准确率,为早期帕金森病语音诊断提供了一种有效的解决思路。
- 25 February 2024
- Vol. 41 (1) , 17-25
基于样本增强的帕金森病识别算法研究
帕金森病患者早期存在声带损伤,其声纹特征与健康人存在明显差异,可以利用该差异识别帕金森病,但帕金森病患者声纹数据样本不足,因此本文提出双自注意力深度卷积生成对抗网络模型进行样本增强,生成高分辨率的语谱图,进而采用深度学习方法进行帕金森病识别。该模型通过增加网络深度并结合梯度惩罚、频谱归一化技术改进样本的纹理清晰度,并且构建一个基于迁移学习的纯粹的卷积神经网络家族(ConvNeXt)作为分类网络,以此提取声纹特征并进行分类,提升了帕金森病识别准确率。在帕金森病语音数据集上进行本文算法有效性验证实验,对比样本增强前,本文所提模型生成的样本清晰度以及弗雷谢起始距离(FID)均得到提高,并且本文网络模型能够获得98.8%的准确率。本文研究结果表明,基于双自注意力深度卷积生成对抗网络样本增强的帕金森病识别算法能够准确区分健康人和帕金森病患者,有助于解决帕金森病早期识别声纹数据样本不足的问题。综上,本文方法有效提高小样本帕金森病语音数据集分类准确率,为早期帕金森病语音诊断提供了一种有效的解决思路。
- 29 April 2023
- Published by Elsevier in Construction and Building Materials
- Vol. 385, 131427
Terahertz Non-destructive testing and imaging of corrosion in coated steel plates
- 29 April 2023
- Published by Elsevier in Construction and Building Materials
- Vol. 385, 131427
Terahertz Non-destructive testing and imaging of corrosion in coated steel plates
- Open Access
- 24 April 2023
- Published by Institution of Engineering and Technology (IET) in IET Signal Processing
- Vol. 17 (4)
Heuristic adaptive threshold detection method for neuronal spikes
In recent years, the development of microelectrode arrays and multichannel recordings has provided opportunities for high‐precision detection in signal processing. The study of neuronal frontal potentials has been rapidly emerging as an important component in brain‐computer interface and neuroscience research. Neuronal spike detection provides a basis for neuronal discharge analysis and nucleus cluster identification; its accuracy depends on feature extraction and classification, which affect neuronal decoding analysis. However, improving the detection accuracy of spike potentials in highly noisy signals remains a problem. IThe authors propose a heuristic adaptive threshold spike‐detection algorithm that removes noise and reduces the phase shift using a zero‐phase Butterworth infinite impulse response filter. Next, heuristic thresholding is applied to obtain spike points, remove repetitions, and achieve robust spike detection. The proposed algorithm achieved an average accuracy of 95.40% using extracellular spiked datasets and effectively detected spikes.
- 24 April 2023
- Published by Institution of Engineering and Technology (IET) in IET Signal Processing
- Vol. 17 (4)
Open AccessHeuristic adaptive threshold detection method for neuronal spikes
In recent years, the development of microelectrode arrays and multichannel recordings has provided opportunities for high‐precision detection in signal processing. The study of neuronal frontal potentials has been rapidly emerging as an important component in brain‐computer interface and neuroscience research. Neuronal spike detection provides a basis for neuronal discharge analysis and nucleus cluster identification; its accuracy depends on feature extraction and classification, which affect neuronal decoding analysis. However, improving the detection accuracy of spike potentials in highly noisy signals remains a problem. IThe authors propose a heuristic adaptive threshold spike‐detection algorithm that removes noise and reduces the phase shift using a zero‐phase Butterworth infinite impulse response filter. Next, heuristic thresholding is applied to obtain spike points, remove repetitions, and achieve robust spike detection. The proposed algorithm achieved an average accuracy of 95.40% using extracellular spiked datasets and effectively detected spikes.
- Open Access
- 25 August 2022
- Published by Springer Nature in Research Square
Study the evaluation model of closed-loop DBS strategies based on the optimal stimulation level
Closed-loop deep brain stimulation (DBS) is currently a research hotspot in the treatment of Parkinson's disease, but a variety of stimulation strategies will increase the selection time and cost in animal experiments and clinical studies. And there is no obvious difference in the stimulation effect between similar strategies, which makes the selection process more redundant. A comprehensive evaluation model based on the optimal level was proposed to select the best strategy in similar strategies. Two similar threshold stimulation strategies were used to demonstrate the screening process. Firstly, the Analytic Hierarchy Process was used to assign the weights of the three indexes of Similar to UPDRS Estimates, β power and energy consumption. Then the values of all indexes were calculated, and the stimulus threshold with the optimal improvement effect was selected according to Similar to UPDRS Estimates. Finally, combined with weights and indexes values, the comprehensive scores were calculated according to the comprehensive evaluation model under the optimal stimulus conditions. By comparing the scores, the optimal stimulus threshold of one strategy was 52%, while the other was 72%. And the stimulus effect of the strategy with the optimal stimulus threshold 72% was better than that of 52%. The results show that the evaluation model based on the optimal stimulus condition can get a unique conclusion, and meet the selection of the optimal strategy between two similar strategies.
- 25 August 2022
- Published by Springer Nature in Research Square
Open AccessStudy the evaluation model of closed-loop DBS strategies based on the optimal stimulation level
Closed-loop deep brain stimulation (DBS) is currently a research hotspot in the treatment of Parkinson's disease, but a variety of stimulation strategies will increase the selection time and cost in animal experiments and clinical studies. And there is no obvious difference in the stimulation effect between similar strategies, which makes the selection process more redundant. A comprehensive evaluation model based on the optimal level was proposed to select the best strategy in similar strategies. Two similar threshold stimulation strategies were used to demonstrate the screening process. Firstly, the Analytic Hierarchy Process was used to assign the weights of the three indexes of Similar to UPDRS Estimates, β power and energy consumption. Then the values of all indexes were calculated, and the stimulus threshold with the optimal improvement effect was selected according to Similar to UPDRS Estimates. Finally, combined with weights and indexes values, the comprehensive scores were calculated according to the comprehensive evaluation model under the optimal stimulus conditions. By comparing the scores, the optimal stimulus threshold of one strategy was 52%, while the other was 72%. And the stimulus effect of the strategy with the optimal stimulus threshold 72% was better than that of 52%. The results show that the evaluation model based on the optimal stimulus condition can get a unique conclusion, and meet the selection of the optimal strategy between two similar strategies.
- 27 March 2022
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 60 (5) , 1225-1237
An improved YOLO Nano model for dorsal hand vein detection system
To date, venipuncture, the most necessary and fundamental medical means, still remains a challenging task for medical stuff due to significant individual differences in vein condition. Thanks to mature development in near-infrared (NIR) imaging technology, a series of venepuncture auxiliary equipment has been devised and put into use. Yet, previous researches concentrated more on vein pattern segmentation, failing to materialize the identification of veins suitable to puncture in an embedded system. Given the above, we propose an approach to detect and locate the optimal veins fully utilizing the state-of-the-art deep learning and image processing technologies in order to provide a more practical reference. Firstly, a dedicated NIR-based puncturable vein positioning system is designed, realizing collection of dorsal hand vein images as well as the rapid and accurate location of veins suitable to puncture. Secondly, considering the limitations of embedded devices on computation ability and memory, an improved network based on YOLO Nano, named YOLO Nano-Vein, is presented with architecture trimmed, output scales reduced, and an atrous spatial pyramid pooling (ASPP) added. Finally, average precision (AP) is increased from 91.68 to 93.23%, and the detection time and parameters of network are reduced by 22% and 17.5%, respectively, which validates the proposed network achieves higher accuracy with less detection time in comparison with YOLO Nano and YOLOv3, indicating stronger applicability for detection tasks on embedded devices. Graphical abstract
- 27 March 2022
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 60 (5) , 1225-1237
An improved YOLO Nano model for dorsal hand vein detection system
To date, venipuncture, the most necessary and fundamental medical means, still remains a challenging task for medical stuff due to significant individual differences in vein condition. Thanks to mature development in near-infrared (NIR) imaging technology, a series of venepuncture auxiliary equipment has been devised and put into use. Yet, previous researches concentrated more on vein pattern segmentation, failing to materialize the identification of veins suitable to puncture in an embedded system. Given the above, we propose an approach to detect and locate the optimal veins fully utilizing the state-of-the-art deep learning and image processing technologies in order to provide a more practical reference. Firstly, a dedicated NIR-based puncturable vein positioning system is designed, realizing collection of dorsal hand vein images as well as the rapid and accurate location of veins suitable to puncture. Secondly, considering the limitations of embedded devices on computation ability and memory, an improved network based on YOLO Nano, named YOLO Nano-Vein, is presented with architecture trimmed, output scales reduced, and an atrous spatial pyramid pooling (ASPP) added. Finally, average precision (AP) is increased from 91.68 to 93.23%, and the detection time and parameters of network are reduced by 22% and 17.5%, respectively, which validates the proposed network achieves higher accuracy with less detection time in comparison with YOLO Nano and YOLOv3, indicating stronger applicability for detection tasks on embedded devices. Graphical abstract