2021 PHAI Seed Grant Awards
Radiomics approaches for elucidating normal tissue toxicity in the brain caused by changes in radiation dose rate using MRI and CBCT imaging
Charles Limoli
The main goal of this project is to provide a discriminating and predictive algorithm capable of elucidating normal tissue toxicities in the brain caused by changes in radiation dose rate using a machine learning analysis of Magnetic Resonance Imaging (MRI) and Cone-Beam Computed Tomography (CBCT) images. A radiomics analysis and a deep learning approach will be implemented in order to investigate the biological mechanisms of the in vivo “FLASH effect”, involving the exposure of mice to ultra-high dose rate whole brain irradiation. The hypothesis is that a radiomics classification algorithm coupled with a neural network analysis will both be able to distinguish FLASH irradiated from conventionally (lower dose rate) irradiated mice using MRI and CBCT imagery and will be able to uncover regional differences in susceptibility between each radiation modality.
Early Detection of New Respiratory Diseases
Tony Givargis
Dr. Givargis and colleagues from the Donald Bren School of Information and Computer Sciences and UCI Health aim to build an edge computing infrastructure for multi‐layered processing of anomalous pulmonary CT scans. They are currently focused on the edge of the network, that is, at the scanning site. Using retrospective data, we have achieved over 92% accuracy detecting lungs with a COVID‐19 infection by using hyperdimensional computing. Additionally, modeling anomalies in medical imaging could translate in further work towards other types of diseases.
Alexandru Nicolau
Alexander Veidenbaum
Victor C. Joe
Neftali Watkinson Medina
Discovering new genetic and epigenetic causes of diseases by detecting outliers with deep learning
Xiaohui Xie
We propose to develop a deep learning-based outlier gene detector to discover genes underlying complex diseases, leveraging multiomics data such as transcriptomics, epigenomics, and DNA sequencing together with clinical health data. With this approach, we search for the outlier genes that make each patient unique in gene expression and/or epigenetics. A preliminary application of our method on amyotrophic lateral sclerosis (ALS) indicates that the proposed approach is promising. If successful, the proposed approach offers a fundamentally new way of analyzing omics data for studying complex diseases, with the potential of discovering new genes and epigenetic changes for personalized patient stratification and precision medicine.
Spatial multi-omics machine learning analysis to elucidate the tumor immune microenvironment
Weian Zhao
We will develop AI-based analysis software that can incorporate deep-learning-based segmentation to reveal the composition of tumor-immune populations, overall immune infiltration, enriched co-occurrence of immune subpopulations, and checkpoint expression. Our technology will uniquely enable direct, highly multiplexing, in situ spatial biomarker profiling in a single round of staining and imaging along with a high-throughput, accurate and easy-to-use AI software to translate multi-omics spatial imaging data to improve clinical outcome. Our technology can be quickly and broadly deployed to enable precision medicine and improve patient care in oncology and other diseases areas including neurological diseases.