Zhipeng Cai

Southeast University

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Chengyu Liu

Southeast University

Jianqing Li

State Key Laboratory of Digital Medical Engineering

Jianqing Li

State Key Laboratory of Digital Medical Engineering

Xiu‐Feng Wan

University of Missouri

Kan Luo

Fujian University of Technology

Guohui Lin

University of Alberta

Xiangyu Zhang

China Agricultural University

Lina Zhao

Southeast University

Xingyao Wang

Southeast University

Yu‐Min Shen

The University of Texas Southwestern Medical Center

Cooperative Institutions

Southeast University

State Key Laboratory of Digital Medical Engineering

Chinese Academy of Sciences

Xiamen University

St. Jude Children's Research Hospital

National Research Centre

Nanjing Medical University

University of Hong Kong

Huazhong University of Science and Technology

Chinese Academy of Medical Sciences & Peking Union Medical College

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

E2PNet: Event to Point Cloud Registration with Spatio-Temporal Representation Learning

arXiv (Cornell University) (2023)

Xiuhong Lin Changjie Qiu Zhipeng Cai Siqi Shen Yu Zang

Event cameras have emerged as a promising vision sensor in recent years due to their unparalleled temporal resolution and dynamic range. While registration of 2D RGB images to 3D point clouds is a long-standing problem in computer vision, no prior work studies 2D-3D registration for event cameras. To this end, we propose E2PNet, the first learning-based method for event-to-point cloud registration. The core of E2PNet is a novel feature representation network called Event-Points-to-Tensor (EP2T), which encodes event data into a 2D grid-shaped feature tensor. This grid-shaped feature enables matured RGB-based frameworks to be easily used for event-to-point cloud registration, without changing hyper-parameters and the training procedure. EP2T treats the event input as spatio-temporal point clouds. Unlike standard 3D learning architectures that treat all dimensions of point clouds equally, the novel sampling and information aggregation modules in EP2T are designed to handle the inhomogeneity of the spatial and temporal dimensions. Experiments on the MVSEC and VECtor datasets demonstrate the superiority of E2PNet over hand-crafted and other learning-based methods. Compared to RGB-based registration, E2PNet is more robust to extreme illumination or fast motion due to the use of event data. Beyond 2D-3D registration, we also show the potential of EP2T for other vision tasks such as flow estimation, event-to-image reconstruction and object recognition. The source code can be found at: https://github.com/Xmu-qcj/E2PNet.

RGB color model

Representation

Feature (linguistics)

Optical Flow

10.48550/arxiv.2311.18433

Cite

Citations (1)

Concepts and applications for influenza antigenic cartography.

PubMed (2011)

Zhipeng Cai Tong Zhang Xiu‐Feng Wan

Influenza antigenic cartography projects influenza antigens into a two or three dimensional map based on immunological datasets, such as hemagglutination inhibition and microneutralization assays. A robust antigenic cartography can facilitate influenza vaccine strain selection since the antigenic map can simplify data interpretation through intuitive antigenic map. However, antigenic cartography construction is not trivial due to the challenging features embedded in the immunological data, such as data incompleteness, high noises, and low reactors. To overcome these challenges, we developed a computational method, temporal Matrix Completion-Multidimensional Scaling (MC-MDS), by adapting the low rank MC concept from the movie recommendation system in Netflix and the MDS method from geographic cartography construction. The application on H3N2 and 2009 pandemic H1N1 influenza A viruses demonstrates that temporal MC-MDS is effective and efficient in constructing influenza antigenic cartography. The web sever is available at http://sysbio.cvm.msstate.edu/AntigenMap .

Diva

Source

Cite

Citations (25)

Graph Convolutional Network with Connectivity Uncertainty for EEG-based Emotion Recognition

arXiv (Cornell University) (2023)

Hongxiang Gao Xiangyao Wang Zhenghua Chen Min Wu Zhipeng Cai

Automatic emotion recognition based on multichannel Electroencephalography (EEG) holds great potential in advancing human-computer interaction. However, several significant challenges persist in existing research on algorithmic emotion recognition. These challenges include the need for a robust model to effectively learn discriminative node attributes over long paths, the exploration of ambiguous topological information in EEG channels and effective frequency bands, and the mapping between intrinsic data qualities and provided labels. To address these challenges, this study introduces the distribution-based uncertainty method to represent spatial dependencies and temporal-spectral relativeness in EEG signals based on Graph Convolutional Network (GCN) architecture that adaptively assigns weights to functional aggregate node features, enabling effective long-path capturing while mitigating over-smoothing phenomena. Moreover, the graph mixup technique is employed to enhance latent connected edges and mitigate noisy label issues. Furthermore, we integrate the uncertainty learning method with deep GCN weights in a one-way learning fashion, termed Connectivity Uncertainty GCN (CU-GCN). We evaluate our approach on two widely used datasets, namely SEED and SEEDIV, for emotion recognition tasks. The experimental results demonstrate the superiority of our methodology over previous methods, yielding positive and significant improvements. Ablation studies confirm the substantial contributions of each component to the overall performance.

Discriminative model

Smoothing

10.48550/arxiv.2310.14165

Cite

Citations (0)

PVC Recognition for Wearable ECGs Using Modified Frequency Slice Wavelet Transform and Convolutional Neural Network

Computing in cardiology (2019)

Zhongyao Zhao Xingyao Wang Zhipeng Cai Jianqing Li Chengyu Liu

Progress in wearable techniques makes the long-term daily electrocardiogram (ECG) monitoring possible.Premature ventricular contraction (PVC) is one of the most common cardiac arrhythmias.This study proposed a method by combining the modified frequency slice wavelet transform (MFSWT) and convolutional neural network (CNN).Training data are from the 2018 China physiological signal challenge (934 PVC and 906 non-PVC recordings).The first 10-s ECG waveforms in each recording were transformed into 2-D time-frequency images (frequency range of 0-50 Hz and size of 300 × 100) using MFSWT.A 25-layer CNN structure was constructed, which includes five convolution layers with kernel size of 3×3, five dropout layers, five ReLU layers, five maximum pooling layers with kernel size of 2 × 2, a flatten layer, two fully connected layers, as well as the input and output layers.Test data were recorded from 12-lead Smart ECG vests, including 775 PVC and 742 non-PVC recordings.Results showed that, the proposed method achieved a high accuracy of 97.89% for PVC/non-PVC episodes classification, indicating that the combination of MFSWT and CNN provides new insight to accurately identify PVC from the wearable ECG recordings.

Continuous wavelet transform

10.22489/cinc.2019.138

Cite

Citations (9)

A Faster R CNN-based Real-time QRS Detector

Computing in cardiology (2019)

Yang Hao Mao Lin Huang Zhipeng Cai Yingjia Yao Chengyu Liu

Accurate QRS location keeps challenging in dynamic electrocardiograms (ECGs).This study addressed this issue and developed a novel faster R convolutional neural network (CNN) model-based real-time QRS detection algorithm.Firstly, ECGs were segmented into 10-s length episodes, and each episode was transformed into a 2-D image with a pixel size of 200200 (VOC2007 format).Labelled QRS location information was used to generate the QRS bounding boxes.A faster R CNN model was constructed.Candidates of QRS bounding boxes were extracted by the region proposal networks (RPN).Then, the boxes with small probabilities were excluded according to the rules of probability distribution and QRS location relationship.Finally, locations of QRS complexes were determined based on the geometric features and threshold rule.The proposed algorithm was trained on the MIT/BIH arrhythmia database and verified on the 24-h wearable ECGs.Five-fold cross validation on 24-h wearable ECG recordings from 20 subjects generated a sensitivity of 98.76%, a positive predictivity of 98.52% and an accuracy of 97.32% compared to the manual annotations.In addition, the cost time of the new algorithm for processing a 10-s ECG episode was less than 20 ms under the experiments of CPU i7-2600 3.40 GHz, 8 GB RAM, tesla M60 GPU and 16 GB graphics memory.

Bounding overwatch

10.22489/cinc.2019.053

Cite

Citations (7)

Robust PVC Identification by Fusing Expert System and Deep Learning

Biosensors (2022)

Zhipeng Cai Tiantian Wang Yu‐Min Shen Yantao Xing Ruqiang Yan

Premature ventricular contraction (PVC) is one of the common ventricular arrhythmias, which may cause stroke or sudden cardiac death. Automatic long-term electrocardiogram (ECG) analysis algorithms could provide diagnosis suggestion and even early warning for physicians. However, they are mutually exclusive in terms of robustness, generalization and low complexity. In this study, a novel PVC recognition algorithm that combines deep learning-based heartbeat template clusterer and expert system-based heartbeat classifier is proposed. A long short-term memory-based auto-encoder (LSTM-AE) network was used to extract features from ECG heartbeats for K-means clustering. Thus, the templates were constructed and determined based on clustering results. Finally, the PVC heartbeats were recognized based on a combination of multiple rules, including template matching and rhythm characteristics. Three quantitative parameters, sensitivity (Se), positive predictive value (P+) and accuracy (ACC), were used to evaluate the performances of the proposed method on the MIT-BIH Arrhythmia database and the St. Petersburg Institute of Cardiological Technics database. Se on the two test databases was 87.51% and 87.92%, respectively; P+ was 92.47% and 93.18%, respectively; and ACC was 98.63% and 97.89%, respectively. The PVC scores on the third China Physiological Signal Challenge 2020 training set and hidden test set were 36,256 and 46,706, respectively, which could rank first in the open-source codes. The results showed that the combination strategy of expert system and deep learning can provide new insights for robust and generalized PVC identification from long-term single-lead ECG recordings.

Heart beat

Robustness

Autoencoder

10.3390/bios12040185

Cite

Citations (9)

Over-fitting suppression training strategies for deep learning-based atrial fibrillation detection

Medical & Biological Engineering & Computing (2021)

Xiangyu Zhang Jianqing Li Zhipeng Cai Zhang Li Zhenghua Chen

Robustness

Training set

Data set

F1 score

10.1007/s11517-020-02292-9

Cite

Citations (47)

A Novel ECG-Derived Respiration Method Combining Frequency-Domain Feature and Interacting Multiple Model Smoother

IEEE Transactions on Biomedical Engineering (2022)

Kejun Dong Li Zhao Cairong Zou Zhipeng Cai Chang Yan

ECG-derived respiration (EDR) is a low-cost and productive means for capturing respiratory activity. In particular, as the primary procedure in some cardiorespiratory-related studies, the quality of EDR is decisive for the performance of subsequent analyses.In this paper, we proposed a novel EDR method based on the feature derived from the first moment (mean frequency) of the power spectrum (FMS). After obtaining the EDR signal from the feature, we introduced the Interacting Multiple Model (IMM) smoother to enhance the similarity of the EDR signal to the reference respiration. The assessment of the approach consisted of two steps: 1) the performance of extracted feature was verified against R-peak misalignment and noise. 2) the enhancement of IMM smoother to EDR waveforms was evaluated based on waveform correlation and respiratory rate estimation. All the assessments were conducted under the Fantasia database and Drivers database.The FMS improved robustness against R peak offsets compared to most established feature-based EDR algorithms, but a slight 5% improvement of waveform correlation against RR interval-based feature under accurate R peaks. The IMM smoother performed similarly with the Kalman filter in the static database but realized the enhancement of some extent of the EDR waveform in the ambulatory database.The proposed method investigated frequency domain mapping of ECG morphological changes caused by respiratory modulation and explained the EDR signal as a non-stationary time series, which provided a direction of better fitting the natural respiration process and enhancing the EDR waveform.

Robustness

Feature (linguistics)

10.1109/tbme.2022.3204764

Cite

Citations (5)

An integrated framework for evaluation on typical ECG-derived respiration waveform extraction and respiration

Computers in Biology and Medicine (2021)

Kejun Dong Li Zhao Zhipeng Cai Yuwen Li Jianqing Li

Respiration rate

10.1016/j.compbiomed.2021.104593

Cite

Citations (4)

Different Sub-bands Assessment of the Resting-state Prefrontal EEG in Depression Patients

Feifei Chen Lulu Zhao Zhipeng Cai Yumin Li Xiaohan Zang

Depression has been the main cause of the global disease burden. Quantitative and accurate diagnosis of depression is urgent in future clinical practice. This study aims to assess the complexity of depression in different frequency bands and the potential of each band in depression recognition. Firstly, the resting-state prefrontal electroencephalogram (PFEEG) signals of 30 depression patients (DP) and 31 healthy controls (HC) were acquired and decomposed into four frequency bands. Then, sample entropy (SEn) and fuzzy entropy (FEn) of each frequency band were calculated and statistically analyzed between DP and HC. Finally, the classification performance of each frequency band in recognizing the two groups was verified. The results illustrated that the complexity of the alpha band in DP showed a significant increasing trend (p<0.01) compared with HC. The best classification performance was achieved in the alpha band using SEn and FEn. Different frequency bands of prefrontal EEG might be a potential and practical biomarker in distinguishing DP and HC.

Depression

Frequency band

10.1109/icsmd57530.2022.10058461

Cite

Citations (0)