Xiang Xinyuan

Dr. Xiang Xinyuan

orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
20 March 2025 in Energies
Verified
Journal Article
Open access
PDF file
SWShuaibo Wang
Shuaibo Wang
XXXiang Xinyuan
Xiang Xinyuan
JZJie Zhang
Jie Zhang
ZLZhuohang Liang
Zhuohang Liang
SLShufang Li
Shufang Li
CWChenguang Wang
Chenguang Wang
Transient stability assessments and state prediction are critical tasks for power system security. The increasing integration of renewable energy sources has introduced significant uncertainties into these tasks. While AI has shown great potential, most existing AI-based approaches focus on single tasks, such as either stability assessments or state prediction, limiting their practical applicability. In power system operations, these two tasks are inherently coupled, as system states directly influence stability conditions. To address these challenges, this paper presents a multi-task learning framework based on spatiotemporal graph convolutional networks that efficiently performs both tasks. The proposed framework employs a spatiotemporal graph convolutional encoder to capture system topology features and integrates a self-attention U-shaped residual decoder to enhance prediction accuracy. Additionally, a Multi-Exit Network branch with confidence-based exit points enables efficient and reliable transient stability assessments. Experimental results on IEEE standard test systems and real-world power grids demonstrate the framework’s superiority as compared to state-of-the-art AI models, achieving a 48.1% reduction in prediction error, a 6.3% improvement in the classification F1 score, and a 52.1% decrease in inference time, offering a robust solution for modern power system monitoring and safety assessments.

10 June 2025 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
ZWZhiyong Wang
Zhiyong Wang
SYSixing Yin
Sixing Yin
YWYining Wang
Yining Wang
WYWenyu Yin
Wenyu Yin
XXXiang Xinyuan
Xiang Xinyuan
SLShufang Li
Shufang Li
Accurate segmentation of the left ventricular myocardium in cardiac CT images is essential for diagnosing and treating cardiovascular diseases. This is considered particularly critical in clinical workflows enabled by the Internet of Medical Things (IoMT), where real-time remote monitoring and rapid analysis of cardiac function are fundamental to timely and effective intervention. Deep learning techniques have achieved significant progress in automatically outlining target objects with reduced human effort. However, most of the deep learning approaches focus on pixel-level loss minimization in model training, and fail to capture the global shape characteristics of anatomical structures, resulting in inconsistent segmentation. In this paper, we propose an end-to-end deep learning framework, which incorporates shape priors derived using kernel principal component analysis (KPCA) to address such limitation. A shape loss function is introduced to enforce anatomical shape consistency by integrating shape prior of myocardial structures during model training. Comprehensive experiments on both convolutional and transformer-based architectures demonstrate that the proposed method significantly enhances segmentation accuracy and anatomical fidelity, particularly in regions critical for diagnosis, such as mitral valve, interventricular septum and papillary muscles. Moreover, we show that integration of learnable Gaussian kernel parameters in KPCA enhances adaptability, outperforming that with fixed kernel parameters. We believe that the improved segmentation accuracy helps doctors make more precise assessments of the heart. This is especially useful in IoMT-based healthcare systems, where reliable segmentation supports quick diagnosis, remote monitoring, and smarter medical decisions, leading to better care and outcomes for patients with heart disease.

18 June 2026 in Journal of Imaging
Verified
Journal Article
Open access
PDF file
XXXiang Xinyuan
Xiang Xinyuan
WYWenyu Yin
Wenyu Yin
JLJiayue Li
Jiayue Li
Pathological complete response (pCR) after neoadjuvant chemotherapy (NACT) provides an endpoint for treatment evaluation in breast cancer. Multi-sequence breast MRI can support pCR prediction, but routine examinations may lack usable T1-weighted or T2-weighted sequences. Many models merge radiomic and deep features by concatenation, leaving the interaction between handcrafted descriptors and learned representations weakly specified. We developed a radiomics-guided framework for pCR prediction from multi-sequence breast MRI. The model uses a multi-branch 2.5D encoder for sequence-specific features, radiomics-guided channel recalibration, and masked token fusion to aggregate available sequence tokens. We evaluated the framework on 157 patients from the I-SPY1 Trial cohort with patient-level five-fold cross-validation, fixed sequence-combination analysis, and slice-window sensitivity analysis. The full model achieved 78.4% accuracy and 0.809 AUC, compared with 75.8% accuracy and 0.788 AUC for the strongest channel-concatenation baseline. In this cohort, radiomics-guided multi-sequence learning was feasible, with external validation required before clinical interpretation.

4 August 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
XXXiang Xinyuan
Xiang Xinyuan
JZJie Zhang
Jie Zhang
JLJiayue Li
Jiayue Li
JZJie Zeng
Jie Zeng
HWHongyu Wang
Hongyu Wang
DZDongxia Zhang
Dongxia Zhang
Transient simulation studies in modern power systems produce dense multi-channel recordings that contain the evidence needed for engineering review, but converting these waveforms into traceable reports remains labor intensive. Directly prompting large language models with raw waveform samples can omit local events, misstate numerical evidence, and weaken stability conclusions. This paper advances a report-oriented formulation of transient recording analysis and presents an evidence-constrained multimodal framework that converts dense simulation outputs into auditable engineering reports. Deterministic programs first compute numerical indicators and rule-based stability evidence, render grouped waveform figures, and segment selected channels to extract local trend, range, and extrema attributes. Multimodal generation then converts channel evidence into textual waveform descriptions, while a final language model writes the report under an evidence-consistency protocol that prioritizes rule evidence over numerical indicators, channel descriptions, and visual figures. We evaluate the framework on 320 BPA simulation cases from the IEEE 9-bus, IEEE 39-bus, IEEE 68-bus, and an anonymized regional 109-bus system. Across the tested systems, the framework preserves deterministic stability categories in all cases, achieves numerical faithfulness of 0.9803--0.9927, trend-description accuracy of 0.9000--0.9750, and expert usability scores of 4.23--4.52. Component analysis shows that deterministic evidence constrains conclusions, channel descriptions support trend interpretation, figures improve reviewability, and the consistency protocol reduces unsupported claims. The results indicate that deterministic transient analysis and LLM-based reporting can be combined for offline simulation review within the tested short-circuit simulation scope.

23 April 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
XXXiang Xinyuan
Xiang Xinyuan
WYWenyu Yin
Wenyu Yin
JLJiayue Li
Jiayue Li
SLShufang Li
Shufang Li
Pathological complete response (pCR) after neoadjuvant therapy is an important indicator of treatment response and prognosis in breast cancer. Multi-modal breast MRI provides complementary information for pCR prediction, but existing methods often assume complete modality availability and do not fully exploit the complementary value of radiomics and deep features. To address these limitations, we propose a radiomics-guided multi-modal learning framework for pCR prediction from breast MRI under incomplete modality settings. The model employs a multi-branch 2.5D encoder to extract modality-specific features, a radiomics-guided gating module to enhance deep representations with handcrafted priors, and a masked fusion strategy to adaptively integrate available modalities while excluding missing ones. Experiments on the I-SPY1 Trial dataset show that the proposed method achieves promising performance and maintains robustness under incomplete modality settings. These results suggest that the proposed framework can effectively integrate multi-modal MRI and radiomics information for pCR prediction and shows potential under incomplete modality settings.