Xiaopu He

Dr. Xiaopu He


Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 May 2023 in Medical Image Analysis
Verified
Journal Article
Closed access
KWKai-Ni Wang
Kai-Ni Wang
SZShuaishuai Zhuang
Shuaishuai Zhuang
QRQi-Yong Ran
Qi-Yong Ran
PZPing Zhou
Ping Zhou
JHJie Hua
Jie Hua
XHXiaopu He
Xiaopu He

24 August 2023 in IEEE Journal of Biomedical and Health Informatics
Verified
Journal Article
Closed access
KWKaini Wang
Kaini Wang
SZShuaishuai Zhuang
Shuaishuai Zhuang
JMJuzheng Miao
Juzheng Miao
YCYang Chen
Yang Chen
JHJie Hua
Jie Hua
SLShuo Li
Shuo Li
The automatic and dependable identification of colonic disease subtypes by colonoscopy is crucial. Once successful, it will facilitate clinically more in-depth disease staging analysis and the formulation of more tailored treatment plans. However, inter-class confusion and brightness imbalance are major obstacles to colon disease subtyping. Notably, the Fourier-based image spectrum, with its distinctive frequency features and brightness insensitivity, offers a potential solution. To effectively leverage its advantages to address the existing challenges, this article proposes a framework capable of thorough learning in the frequency domain based on four core designs: the position consistency module, the high-frequency self-supervised module, the complex number arithmetic model, and the feature anti-aliasing module. The position consistency module enables the generation of spectra that preserve local and positional information while compressing the spectral data range to improve training stability. Through band masking and supervision, the high-frequency autoencoder module guides the network to learn useful frequency features selectively. The proposed complex number arithmetic model allows direct spectral training while avoiding the loss of phase information caused by current general-purpose real-valued operations. The feature anti-aliasing module embeds filters in the model to prevent spectral aliasing caused by down-sampling and improve performance. Experiments are performed on the collected five-class dataset, which contains 4591 colorectal endoscopic images. The outcomes show that our proposed method produces state-of-the-art results with an accuracy rate of 89.82%. The source code is released at https://github.com/soleilssss/AFACNet .

3 October 2025 in Medicine
Verified
Journal Article
Open access
QLQinggui Li
Qinggui Li
XWXiao Wang
Xiao Wang
QYQian Ye
Qian Ye
LLLiping Li
Liping Li
XHXiaopu He
Xiaopu He
LLLingling Liu
Lingling Liu
This study aims to develop a prediction model based on machine learning algorithms to predict the risk of stroke by analyzing physical activity and other risk factors. We conducted a retrospective analysis of 134 stroke patients treated at the Department of Neurology, the First Affiliated Hospital of Nanjing Medical University, from July 1, 2021, to May 31, 2023, and 354 non-stroke individuals recruited from the Fenghuang Community Health Screening Program in Nanjing during the same period. Eight machine learning models, including extreme gradient boosting, support vector machine, random forest (RF), neural network, Naive Bayesian, logistic regression, K-nearest neighbor, and decision tree, were used to build the prediction models. Variables were selected using the least absolute shrinkage and selection operator and the multivariable logistic regression analysis. The models were evaluated using receiver operating characteristic (ROC) curves, area under the ROC curve, precision-recall (PR) curves, area under the PR curve, accuracy, sensitivity, specificity, and precision. Shapley Additive Explanations were employed to determine feature importance. The results demonstrated that the RF algorithm performed well in terms of area under the ROC curve (0.96), area under the PR curve (0.92), specificity (0.97), and precision (0.92). Shapley Additive Explanations analysis revealed that the number of weekly exercise days had the most significant impact on stroke risk, followed by calf circumference, past medical history, gender, body mass index, and Strength, Assistance with walking, Rising from a chair, Climbing stairs, and Falls score. The RF algorithm demonstrated strong predictive performance for stroke risk and may guide clinical decision-making.

12 March 2026 in Frontiers in Medicine
Verified
Journal Article
Open access
PDF file
XWXiao Wang
Xiao Wang
LPLiangpeng Pu
Liangpeng Pu
SYShanshan Yan
Shanshan Yan
SZShuaishuai Zhuang
Shuaishuai Zhuang
JHJie Hua
Jie Hua
XHXiaopu He
Xiaopu He
Background: Gastric subepithelial lesions (SELs) are typically covered by intact mucosa, which challenges the determination of their layer of origin under conventional endoscopy. Endoscopic ultrasonography (EUS) is indispensable for diagnosing SELs. However, EUS operation and interpretation are more challenging than standard endoscopy and are subject to inter-observer variability. This study aimed to develop a novel AI system based on the MedMamba architecture to assist clinicians in identifying the layer of origin of gastric SELs. Methods: We retrospectively collected data from patients who underwent EUS at the First Affiliated Hospital of Nanjing Medical University between May 1, 2016 and May 1, 2023. The dataset comprised 1,855 images from 320 patients. Images were split into training, validation, and test sets at an 8:1:1 ratio at the patient level to prevent data leakage. A structured State space sequence model (MedMamba) was trained and the performance was compared against endoscopists. Results: The proposed MedMamba model achieved an overall accuracy of 92.04% (95%CI: 90.33–93.75%), with 94.83% (95%CI: 93.22–96.44%) specificity and 81.11% (95%CI: 75.19–87.03%) sensitivity in the five-category classification. It achieved high sensitivity and accuracy, outperforming other AI models. Conclusion: The MedMamba-based AI system demonstrated superior performance in discriminating the layer of origin of SELs compared with other AI models, indicating its potential utility in reducing diagnostic variability and enhancing clinical diagnostic workflows.

6 March 2026 in Frontiers in Medicine
Verified
Journal Article
Open access
PDF file
LPLiangpeng Pu
Liangpeng Pu
XWXiao Wang
Xiao Wang
SYShanshan Yan
Shanshan Yan
SZShuaishuai Zhuang
Shuaishuai Zhuang
XHXiaopu He
Xiaopu He
Objective: To develop a convolutional neural network (CNN) framework for the automated segmentation and classification of esophageal lesions in endoscopic images. Methods: (1) Lesion localization was performed using a Region-based Convolutional Neural Network (R-CNN). (2) A dual-stream Esophageal Lesion Network (ELNet) was developed to classify images into four diagnostic categories. (3) Lesion segmentation was carried out using an ensemble of three U-Net architectures. Results: The dual-stream ELNet achieved a classification accuracy of 92.14%, with 97.1% specificity and 88.74% sensitivity. The segmentation module based on U-Net attained an overall accuracy of 95.54% and a lesion segmentation sensitivity of 82.89%. The dual-stream ELNet consistently outperformed single-stream baseline networks, and the integrated segmentation-with-classification architecture demonstrated enhanced adaptability across diverse lesion types. Conclusion: The proposed CNN framework enables accurate, robust, and simultaneous classification and segmentation of esophageal endoscopic lesions, exhibiting high performance and clinical potential.