A photo of Lili He.

Lili He, PhD

  • Associate Professor, UC Department of Radiology



I am a computer scientist and expert in magnetic resonance imaging (MRI) with a long-standing commitment to develop and validate robust, clinically-effective computer-aided diagnosis systems. My research areas include machine learning, deep learning and medical imaging. I am committed to lending my expertise in neuroimaging and computer science to facilitate major breakthroughs in the medical field by optimizing imaging acquisition and aiding doctors in disease diagnosis, outcome prediction, image segmentation and interpretation as well as treatment decision-making and assessment.

I have been leading our team of artificial intelligence (AI) for computer aided diagnosis (AI-CAD) to develop algorithms to:

  1. Improve image quality. Image quality directly influences diagnostic quality of the examination by characterizing tissue compared to another and, ultimately, highlighting morphological or functional abnormality. We have developed novel and practical imaging optimization techniques to achieve high image quality, especially by focusing on optimizing the trade-offs between image quality (signal-to-noise ratio, contrast-to-noise-ratio, spatial resolution) and scan time.
  2. Establish automate image processing pipeline. It has been a challenging task to bring data from multimodal MR images, together into a cohesive framework to enable cross-subject comparisons and multi-modal analysis. We have established neonatal anatomical and functional MRI automatic processing pipelines to minimize human efforts.
  3. Detect pathology. Medical image segmentation separates structures of interests from the background and from each other. It plays an important role in diseases’ pattern recognition for further diagnosis, monitoring and treatment. We have reported a sequential development of objective methods to quantitatively diagnose diffuse matter abnormalities and other injuries.
  4. Identify patients at risk of disease/adverse outcome and identify novel imaging biomarkers. We have been developing clinically relevant disease diagnosis and risk prediction AI models for neurodevelopmental deficits, attention deficit hyperactivity disorder and autism, which will allow more frequent disease monitoring to assess treatment response/disease progression, and potentially improve overall outcomes and lower healthcare costs. Ultimately, the developed AI models will enhance our abilities to diagnose disease in a quantitative, noninvasive, patient-friendly manner as well as to provide more patient-centric, precision medicine.

The current AI technique is rapidly moving from an experimental phase to an implementation phase in many fields. It is expected that medical AI will surpass human performance in specific applications within the coming years. Physicians and patients will likely benefit from the human-AI interaction. Since I have distinguished myself as a front-runner in medical imaging-based AI, I look forward to leading this endeavor at Cincinnati Children's Hospital Medical Center.

BS: Electrical Engineering, Tsinghua University, Beijing, China, 1998.

MS: Computer Science, University of Missouri, Columbia, MO, 2003.

PhD: Computer Science and Engineering, University of Connecticut, Storrs, CT, 2008.

Post-Doc: Massachusetts General Hospital, Harvard Medical School, Boston, MA, 2010.


Machine learning; deep learning; medical image processing and analysis

Research Areas

Imaging, Developmental Biology, Neonatology


A novel collaborative self-supervised learning method for radiomic data. Li, Z; Li, H; Ralescu, AL; Dillman, JR; Parikh, NA; He, L. Neuroimage. 2023; 277:120229.

Dynamic weighted hypergraph convolutional network for brain functional connectome analysis. Wang, J; Li, H; Qu, G; Cecil, KM; Dillman, JR; Parikh, NA; He, L. Medical Image Analysis. 2023; 87:102828.

Recent Advances in Explainable Artificial Intelligence for Magnetic Resonance Imaging. Qian, J; Li, H; Wang, J; He, L. Diagnostics. 2023; 13:1571.

A Semi-Supervised Graph Convolutional Network for Early Prediction of Motor Abnormalities in Very Preterm Infants. Li, H; Li, Z; Du, K; Zhu, Y; Parikh, NA; He, L. Diagnostics. 2023; 13:1508.

PREDICTION OF FONTAN OUTCOMES USING T2-WEIGHTED MRI RADIOMIC FEATURES AND MACHINE LEARNING. Prasad, A; Dillman, J; Lubert, A; Trout, A; He, L; Li, H. Journal of the American College of Cardiology. 2023; 81:1618.

Prenatal tobacco smoke exposure and risk of brain abnormalities on magnetic resonance imaging at term in infants born very preterm. Mahabee-Gittens, EM; Kline-Fath, BM; Harun, N; Folger, AT; He, L; Parikh, NA. 2023; 5:100856.

A Novel Collaborative Self-Supervised Learning Method for Radiomic Data. Li, Z; Li, H; Ralescu, AL; Dillman, JR; Parikh, NA; He, L. 2023; abs/2302.09807.

Diffuse excessive high signal intensity in the preterm brain on advanced MRI represents widespread neuropathology. Kline, JE; Dudley, J; Illapani, VS P; Li, H; Kline-Fath, B; Tkach, J; He, L; Yuan, W; Parikh, NA. Neuroimage. 2022; 264:119727.

A novel Ontology-guided Attribute Partitioning ensemble learning model for early prediction of cognitive deficits using quantitative Structural MRI in very preterm infants. Li, Z; Li, H; Braimah, A; Dillman, JR; Parikh, NA; He, L. Neuroimage. 2022; 260:119484.

Current and emerging artificial intelligence applications for pediatric abdominal imaging. Dillman, JR; Somasundaram, E; Brady, SL; He, L. Pediatric Radiology: roentgenology, nuclear medicine, ultrasonics, CT, MRI. 2022; 52:2139-2148.

From the Blog