Artificial intelligence (AI) is increasingly transforming cancer research, and a wide range of AI initiatives are underway at the UF Health Cancer Center. From population sciences to lung cancer screening to clinical trial enrollment, Cancer Center members are leveraging AI in powerful ways.
In 2021, the Cancer Center launched a Cancer AI Working Group to develop collaborative research, expertise and network capacity in AI. The group’s mission is to capitalize on and grow AI capacity at UF to advance AI applications in basic and translational cancer research, bridge interdisciplinary expertise and foster early-career investigators.
Working group co-chairs Qing Lu, Ph.D., professor in the department of biostatistics, and Mattia Prosperi, Ph.D., professor and college coordinator in the department of epidemiology, bring a wealth of expertise and experience in data science. The working group is run by the Biostatistics and Quantitative Sciences Shared Resource in the Division of Quantitative Sciences, which is directed by Ji-Hyun Lee, DrPH, professor in the department of biostatistics, and the Cancer Informatics Shared Resource at the UF Health Cancer Center. The Working Group Committee comprises UF Health Cancer Center members, several of whom hold leadership positions at UF Health.
Since its creation, the working group has recruited several members and held monthly meetings featuring a variety of activities. These have included roundtables with members to discuss research challenges in which AI approaches could be practical, seminars with UF faculty and nationally recognized external speakers, symposia and pilot projects.
Notably, the working group spearheaded the inaugural AI Day for Cancer Research in September 2021. This mini-symposium included both UF and external speakers and early-career scientists. The event was a great success, highlighting current AI-powered cancer research and providing practical information for researchers interested in using the fast-developing technology.
In November 2021, the working group also joined in hosting a cancer-focused set of AI talks as part of the Data Intelligence Symposium (DAISY) organized by UF under the AI initiative. Additionally, in spring 2022, the working group released a request for applications to fund cancer AI-focused pilot projects, and in summer 2022, the first pilot was released. The pilot award went to Kiley Graim, Ph.D., and Mei He, Ph.D., for a project titled “AI-Guided Peptide Design for Precision Cancer Immunotherapy.”
Current AI initiatives now center on aligning AI initiatives with the UF Health Cancer Center’s strategic plan, Momentum 2027. The strategic plan includes several goals, one of which is to conduct transdisciplinary research that leverages expertise and initiatives across the University of Florida. Under this goal, there are two strategies relevant to AI:
- To investigate biological, behavioral, and social determinants of health related to cancer incidence and mortality. This will be accomplished by using integrative approaches, including AI, combining “big data” from the electronic health record, cancer registry data and biospecimens to elucidate mechanisms of carcinogenesis and inform interventional prevention strategies.
- To grow infrastructure to enhance transdisciplinary research initiatives. This will be accomplished by recruiting AI experts to analyze radiomics, biomarkers, multi-omics and electronic health records.
On Sept. 12, the Second Annual AI in Cancer Research mini-symposium will focus on integrative approaches to using AI and using imaging data in cancer research.
Below is a more detailed look at some of the ongoing AI projects that Cancer Center members are working on.
UF Health Cancer Center Imaging Research Group
Personalizing drug delivery in organ transplant
Research Professor, Department of Biostatistics
Associate Professor, Department of Surgery
The adverse effects of over-immunosuppression or under-immunosuppression on posttransplant outcomes present a major challenge in organ transplantation. While under-immunosuppression can increase the risk of rejection, over-immunosuppression increases the risks of cancer development, infection and multiorgan toxicity. Immunosuppressants have a narrow therapeutic index and wide pharmacokinetic variability between patients. Rich patient-specific data including clinical data, genomics, radiomics and biomarkers empower clinicians to precisely predict a patient’s response to drug treatment and identify optimal drug and dose combinations. However, the current tools are limited in their ability to rapidly optimize immunosuppression drug dosing.
In this project, the Cancer Center Imaging Research Group aims to develop an artificial intelligence-based and machine learning-based computing tool that integrates multidomain, multimodal patient data. This tool will proactively guide and individualize immunosuppression in solid organ transplantation. Ultimately, the project aims to minimize toxicity and adverse events and improve graft and patient survival after transplantation.
“This project has the potential to make a meaningful difference in the lives of patients who undergo transplants by using AI to personalize drug delivery.”XiangYang “George” Lou, Ph.D.
Improving diagnostic accuracy in prostate cancer
Assistant Professor, Department of Biostatistics
Assistant Professor, Department of Urology
Arkaprava Roy, Ph.D., is working with Wayne Brisbane, M.D., and his team on a project to improve prostate cancer treatment by using micro-ultrasound imaging.
Prostate cancer is the second most frequently diagnosed cancer in men and the fifth leading cause of death worldwide. The American Cancer Society estimates that there will be 268,490 new prostate cancer cases and 34,500 deaths due to prostate cancer in 2022 in the United States alone. Prostate biopsy is the mainstay of cancer diagnosis and subsequent curative treatment. However, false negatives from biopsies are a significant issue in current prostate biopsy practice. Although MRI-guided prostate biopsy improves cancer detection, it still may miss up to 30% of cancer and adds considerable cost and time to diagnosis.
To improve diagnostic accuracy in prostate cancer, Brisbane and his group use micro-ultrasound imaging and propose implementing AI technology to develop a new targeted biopsy. The current project is blending the expertise of the UF Health Cancer Imaging Research Group and an oncology investigator to develop an efficient AI-based tool to identify cancer lesions and suitable areas for targeted biopsy based on patient-specific systematic biopsy reports, imaging and other relevant biomarkers. With advanced AI machinery, the group hopes to identify cancer lesions by using imaging and other biomarkers, thereby improving biopsy accuracy. The team is also focused on analyzing the efficacy of micro-ultrasound imaging and exploring its potential as an alternative to expensive MRI.
“With the power of advanced technological tools, it is now possible to collect a wide range of medical data very quickly and efficiently. AI tools will be very useful to process and analyze such complex structured high-dimensional data, which has the potential to improve the current state-of-the-art treatments for cancer patients.”Arkaprava Roy, Ph.D.
Cancer AI Working Group Leaders
Professor, Department of Biostatistics; Co-Chair, Cancer AI Working Group
Qing Lu, Ph.D., and his group study the theoretical foundation of neural networks. The team develops new neural network-based methods (e.g., neural-network transformation models and expectile neural networks) for high-dimensional genomic data analysis.
Based on these works, the group received a University of Florida Artificial Intelligence Research Catalyst Fund award for a project, titled “A Kernel Neural Network for High-dimensional Genomic Risk Prediction.”
His group has also published/filed two AI patents (U.S. Patent Application Publication US-2021-0313065-A1 and U.S. Provisional Patent Application Serial No. 63/124,981).
Professor and College Coordinator of AI, Department of Epidemiology; Co-Chair, Cancer AI Working Group
Mattia Prosperi, Ph.D., works in data science, artificial intelligence and biomedical modeling. His research group works on developing original algorithms and applications, using machine learning with a critical eye on causality and designing usable tools. He combines his computer science engineering background with his epidemiology experience to successfully exploit a layered big data analytics paradigm, which integrates multiple domains, such as sociodemographic, ecological, clinical and pathogen sequencing.
Dr. Prosperi serves as the college coordinator of AI and leads the Data Intelligence Systems Lab, promoting interdisciplinary team science, education and scholarly activities. His vision is pursuing ideas in humanely ethical AI for health and beyond.
Examples of AI applications at the Cancer Center
Assessing lung cancer screening
Professor, Department of Biomedical Informatics; Chief Data Scientist, UF Health; Co-Chair, Cancer AI Working Group
Associate Professor, Department of Health Outcomes & Biomedical Informatics
Jiang Bian, Ph.D., and Yi Guo, Ph.D., are working on a project funded by a National Institutes of Health/National Cancer Institute R01 grant to assess the benefits and harms of lung cancer screening in Florida. Lung cancer is the leading cause of cancer-related death in both men and women in the United States. Approximately 70% of lung cancer patients are diagnosed at advanced stages, and the five-year survival rate of advanced stage lung cancer is only 16%.
Low-dose computed tomography is a promising technique for reducing the lung cancer burden, but there are concerns about false positives, the invasiveness of the procedure, postprocedural complications and health care costs. In this study, the investigators are seeking to understand the use of lung cancer screening and associated health care outcomes and costs by using real-world data.
The group aims to develop an innovative computable phenotype algorithm to identify high- and low-risk individuals for lung cancer screening from electronic health record data. They will develop advanced natural language processing methods to extract clinical information related to lung cancer screening from clinical notes, such as radiology reports.
Furthermore, the group aims to develop and validate a microsimulation model of the clinical course of lung cancer screening that incorporates real-world data. This model would be used to estimate the long-term benefits and the cost-effectiveness of lung cancer screening.
Reducing data biases
Assistant Professor, Department of Computer and Information Science and Engineering
Kiley Graim, Ph.D., and her team are developing AI models to understand the molecular effects underlying disease development, progression and response to treatment. The primary focus is human cancer, but the group also does research in several other systems including sepsis and equitable AI.
For example, the group is currently developing an AI method that helps distinguish between traits related to ancestry and traits related to disease. Although current AI disease models have demonstrated their power in identifying disease, many are less accurate in minority populations because those individuals are underrepresented in the entire scientific pipeline (data creation, subtyping, treatment development and disease detection/treatment), Graim said.
From the lab to the clinic, unintentional biases in data can have a negative impact that the group aims to eliminate. By stratifying ancestry-specific effects, the team’s method seeks to create more equitable molecular disease modeling and to eventually improve disease outcomes for all.
“I most enjoy seeing my lab’s models cut through the noise in large molecular data to identify essential disease factors, making a difference in how diseases are identified and treated. AI allows us to integrate big data, enabling us to build more robust models to better understand diseases’ molecular drivers. This allows us to get to the root causes of diseases like cancer, create more equitable AI models and explore more creative approaches to science.”Kiley Graim, Ph.D.
Increasing clinical trial recruitment
Professor, Department of Computer & Information Sciences & Engineering
Benjamin Lok, Ph.D., and his team are levering a variety of AI techniques in ongoing projects that aim to increase recruitment into clinical trials (NIH R24; PIs: Janice Krieger, Ph.D., and Stephen Anton, Ph.D.) and to increase rural and minority screening for colorectal cancer (NIH R01; PI: Janice Krieger, Ph.D.). The researchers are using AI to identify how to best communicate to the public about the importance of clinical trials and complying with screening guidelines for colorectal cancer.
“I’m most excited about helping our Cancer Center experts amplify their intelligence and expertise through insights that AI can help identify. For example, consider how much more impactful the Cancer Center’s health messages will be in improving outcomes if we learn how to best communicate to each of us as individuals.”Benjamin Lok, Ph.D.
Understanding medical data
Associate Professor, Director of Natural Language Processing, Department of Health Outcomes & Biomedical Informatics
Yonghui Wu, Ph.D., is working on several AI-related projects, including:
- A clinical language AI model called GatorTron™, which is the largest AI model in the clinical domain. GatorTron™ can be used to understand medical language in clinical notes (including those of cancer patients). Learn more about the GatorTron™ project.
- A project funded by the Patient-Centered Outcomes Research Institute (PCORI ME-2018C3-14754) to extract social determinants of health (e.g., employment) and study how these social determinants of health could affect the risk of lung, breast and colorectal cancer.
- As part of UF Health’s AI initiative with NVIDIA, a project to develop an AI system that can understand clinical trials and electronic health records and identify patients who meet the recruitment requirements. This project aims to speed up patient recruitment and reduce the human burden.
“If we can enable machines to have intelligence to understand medical data, physicians will be able to leverage both data-driven evidence and their professional knowledge, thus improving healthcare quality.”Yonghui Wu, Ph.D.
Exploring drug design
Professor and Nicholas Bodor Professor In Drug Discovery, Department of Medicinal Chemistry
Chenglong Li, Ph.D., and his team are developing an AI-based drug design platform combining graph-based molecular generators and deep learning-based protein/ligand affinity predictors. The team is also applying AI-based drug design approaches to specific targets, such as IL-6, STAT3, PRMT5 and YAP/TEAD.
“The potential for AI is enormous as the approaches can explore the wider drug-chemical space as never before. Drug/target-binding prediction can complement traditional physics-based strategy. It is absolutely exciting to use the new approaches to design cancer drugs more effectively, for both small-molecule drugs and protein-based therapeutics.”Chenglong Li, Ph.D.