Liangyu Ji

Nanjing Medical University

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Xiangxun Chen

First Affiliated Hospital of Anhui Medical University

Ran Tang

Chengdu University of Information Technology

Meng Zhao

Beijing Institute of Technology

Ge Wu

Southeast University

Xuan Zhang

Nanjing Medical University

Liquan Chen

Chang'an University

Fei‐Yun Wu

Nanjing Medical University

Shuheng Zhang

Inner Mongolia University of Science and Technology

Xuan Zhang

Chinese Academy of Medical Sciences & Peking Union Medical College

Meng Zhao

Xi'an University of Science and Technology

Cooperative Institutions

Nanjing Medical University

Jiangsu Province Hospital

Southeast University

Purple Mountain Laboratories

Southeast University

Nanjing University of Science and Technology

Zhejiang University

United Imaging Healthcare (China)

Chinese Academy of Medical Sciences & Peking Union Medical College

First Affiliated Hospital of Anhui Medical University

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

ERGA

Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering (2021)

Liangyu Ji Tian Yao Ge Wu Liquan Chen Zhongyuan Qin

In recent years, with the rapid development of deep neural networks in the field of artificial intelligence, some of the security issues involved have gradually attracted attention in the industry, one of which is adversarial sample attacks. The attacker inputs carefully designed adversarial samples to the deep learning model, causing the attacked model to output misclassification results with high confidence, which seriously threatens the robustness of the deep learning model. Based on the commonly used deep learning network model, combined with the interpretability research of neural networks, we propose an effective region generation algorithm (ERGA) for adversarial sample generation, which can overcome the defects of the current commonly used algorithms. In our approach, the effective region selection step is added in the adversarial sample generation process, which overcomes the limitation of common adversarial sample generation algorithms that are limited to global pixel perturbation. We try to limit the number of pixels to be changed while maintaining a higher attack success. The algorithm also optimizes the process of counter disturbance generation, solves the uncertainty of the gradient update direction and amplitude in the iterative process. In addition, it also introduces the interpretability research of counter samples, which can be used to a certain extent in the deep learning network. At the same time, ERGA can improve the ability of supervision and self-examination of the classification results.

Interpretability

Robustness

Deep Neural Networks

Sample (material)

10.1145/3501409.3501611

Cite

Citations (0)

Application of T1-weighted and augmented T1-weighted images of multi-parametric MR sequence in detecting neonatal punctate white matter lesions

Magnetic Resonance Imaging (2024)

Liangyu Ji Lingnan Kong Xuan Zhang Xiangxun Chen Chao Lu

Sequence (biology)

White (mutation)

10.1016/j.mri.2024.110317

Cite

Citations (0)

Evaluation of deep gray matter for early brain development usingmulti-parametric MR

Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition (2024)

Xuan Zhang Liangyu Ji Ye Zhang Lin Zhang Ran Tang

Motivation: Few research has reported the differences of the T1, T2* and proton density (PD) values of multi-parametric magnetic resonance (MR) between healthy preterm and term neonates. Goal(s): To evaluate T1, T2* and PD mapping of preterm and term neonate in the deep gray matter. Approach: We performed MRI on a 3.0T MR scanner (Omega, United Imaging Health Care) using multi-parametric MR imaging with flexible design (MULTIPLEX) transverse axis sequence. Results: T1 relaxation time showed significant differences in bilateral putamen (PT) and pallidus (GP) between preterm and term newborns and was negative correlated with gestational age (GA). Impact: Our study demonstrated that the T1 relaxation time in deep gray matter nuclei decreased with GA, and the preterm showed higher T1 relaxation values than the term neonate in PT and GP at term equivalent age.

Gray (unit)

10.58530/2024/2557

Cite

Citations (0)