• marketing@scepscor.org>

Dr. Nianyi Li Research Focus.

Dr. Nianyi Li Research Focus
Assistant Professor at School of Computing
Clemson University

Dr. Nianyi Li is an Assistant Professor in the School of Computing at Clemson University, specializing in machine learning, computer vision, computational photography, and medical image processing. She received her Ph.D. in Computer and Information Sciences from the University of Delaware and has conducted postdoctoral research at Duke University’s Carl E. Ravin Advanced Imaging Laboratories and Louisiana State University’s IVLAB. Her expertise spans deep learning, feature extraction, and computational imaging, with a strong emphasis on developing AI-driven solutions for biomedical applications. As a key researcher in the NSF EPSCoR-funded ADAPT in SC project, Dr. Li is leading efforts to enhance Explainable AI (XAI) frameworks for biomedical devices, ensuring that AI-driven healthcare solutions are both reliable and interpretable.

Her research within the Biomedical AI Core focuses on bridging the gap between AI accuracy and interpretability, addressing fundamental challenges in biomedical imaging and decision support systems. Her work emphasizes three critical areas: 1) integrating XAI techniques with small, unannotated biomedical datasets to improve model performance in data-scarce environments; 2) developing AI methodologies capable of handling multi-modal, high-dimensional clinical data from diverse biomedical devices; and 3) ensuring transparency in AI-driven decision-making processes, such as severity assessment from imaging data. By combining deep learning with advanced feature extraction techniques, she is creating robust, sensor-agnostic AI models that improve diagnostic, prognostic, and therapeutic capabilities. Through her contributions to ADAPT in SC, Dr. Li is helping to position South Carolina as a leader in the development of AI-enabled biomedical technologies, advancing the state’s research capacity and fostering innovation in personalized healthcare solutions.

Unsupervised Deep Learning for Data Restoration

Biomedical imaging techniques such as microscopy, MRI, CT, and X-ray often suffer from noise, distortions, and artifacts, which can compromise diagnostic accuracy and research outcomes. However, due to privacy concerns, high annotation costs, and the specialized nature of medical imaging, acquiring large, high-quality labeled datasets is often impractical. Traditional supervised learning methods struggle in these data-limited scenarios, leading to overfitting and poor generalization across different imaging modalities. Dr. Li’s research, in collaboration with the Medical University of South Carolina (MUSC), focuses on unsupervised deep learning for biomedical data restoration, addressing a critical challenge in medical imaging: the reliance on large, annotated datasets for effective AI-based image enhancement.

Preliminary experiments using an unsupervised self-supervised neural network have shown promising results in biomedical video denoising, particularly in one-photon calcium imaging datasets. The model outperforms state-of-the-art methods in denoising real and synthetic biomedical data containing Gaussian, Poisson, and Impulse noise patterns, demonstrating its robustness and effectiveness in real-world scenarios.

Physics-Informed Deep Learning

Physics-informed deep learning plays a crucial role in enhancing Explainable AI for biomedical imaging by incorporating domain-specific knowledge into AI models, ensuring that predictions align with real-world physical and biological principles. Traditional deep learning approaches often function as “black boxes,” making it difficult to interpret their decision-making processes, especially in critical applications such as medical diagnostics. Dr. Li’s research, funded by NSF RI program, leverages physics-informed deep learning to enhance image restoration and scene understanding in complex visual environments. In the context of ADAPT in SC, Dr. Li extends these techniques to biomedical imaging, where physics-informed DL enhances explainability by integrating biological and imaging physics principles into AI-driven diagnostic and treatment planning tools. This research is particularly valuable for multi-modal biomedical image restoration, severity assessment, and decision support systems, ensuring AI-driven biomedical devices provide interpretable, trustworthy, and clinically relevant insights.