Peirong Liu
Harvard University
pliu17@mgh.harvard.edu
Bio
Peirong Liu is a postdoctoral researcher at Harvard Medical School and Massachusetts General Hospital, working with Dr. Juan Eugenio Iglesias. She obtained her Ph.D. in Computer Science from UNC-Chapel Hill in 2023, supervised by Dr. Marc Niethammer, and did internships at Meta AI. Her research focuses on advancing theory and developing practical algorithms for robust and interpretable models that contribute to reliable and accessible healthcare. Her work has been published in top venues such as CVPR, ICCV, ECCV, NeurIPS, IEEE TMI, MICCAI, and IPMI. Her research has been selected as oral presentations at CVPR, MICCAI, IPMI, and she has been invited for talks at Harvard University, Cornell University, HKUST. She is dedicated to developing robust, generalizable algorithms that deliver long-term benefits to global healthcare systems.
Areas of Research
- AI for Healthcare and Life Sciences
Robust and Interpretable Learning for Modern Healthcare: From Theory to Practice
The unprecedented advances of modern machine learning offer the potential for faster and more accurate data-driven analysis. However, ideal algorithmic setups often fall short in practice, particularly in the diverse healthcare environments. Therefore, the successful deployment of any approach depends on both model and data: theoretical foundations ensure methodological soundness and enhance the model’s applicability; Meanwhile, to achieve generalizable representations, data-driven models must rely on extensive, inclusive data. My research aims to develop robust and adaptable algorithms capable of effectively handling complex and imperfect real-world data, and apply them to unlock new capabilities for enhancing modern healthcare systems. I have developed a series of physics-driven learning frameworks for reliably modeling spatiotemporal dynamics. These frameworks enable continuous dynamics imaging, significantly improving stroke diagnosis and lesion detection. Additionally, I proposed novel modality-agnostic representation learning approaches for brain imaging, leveraging synthetic data to create robust and adaptable foundation models. This work opens new possibilities for expanding access to affordable and portable MR diagnosis. By integrating theory, algorithms, and applications, my research enhances the resilience of machine learning algorithms against the inherent imperfections in real-world data, and ultimately contributes to a safer, more reliable, and accessible healthcare environment.