Research

The accelerated pace by which biomedical big data is being generated in the healthcare enterprise holds enormous promise in developing individualized diagnostic and therapeutic strategies towards precision medicine. By combining genotypic and phenotypic information, new patterns and predictive markers can be uncovered and translated into clinical tools improving patient care. However, the size, complexity, and heterogeneity of the data pose critical barriers in discovering markers that are reproducible and clinically meaningful.

My research contributions have been focused in four areas:

  1. Adapting and validating novel decision models and uncertainty analysis to discover optimal strategies for individual patients;
  2. Formalizing literature for treatment selection and experiment planning;
  3. Improving the generalizability of predictive models; and
  4. Developing translational applications that assist clinical and translational researchers with interpreting available evidence for precision medicine.

These developments are being practically implemented in a variety of clinical domains, including lung cancer, brain cancer (glioblastoma multiforme), prostate cancer, and breast cancer.

I am interested in adapting and validating novel algorithms, translating them to applications that enable precision medicine. My students and I work on problems related to data wrangling, knowledge representation, modeling, and interpretation. We utilize a wide spectrum of approaches from logistic regression and statistical approaches to sequential pattern mining, deep learning, multi-arm/contextual bandits, depending on the problem at hand. While the approaches and methodologies that we develop have broad applicability across the biomedical domain, an emphasis of our work is on enhancing the role of imaging in diagnosis and patient management.

My research is highly multidisciplinary. My students and I collaborate with faculty members from centers across UCLA such as the Center for Domain Specific Computing, Integrative Center for Learning & Memory, Institute for the Risk Sciences, and the Center for Computer Vision and Imaging Biomarkers. We also are engaging in collaborative efforts with industrial partners such as Siemens Healthineers and ProSocial Applications.

Interests

  • Enabling the integrative analysis of clinical, imaging, and genomic data
  • Discovering optimal strategies for screening and therapy using online learning
  • Incorporating uncertainty analysis into clinical decision support tools
  • Utilizing simulation methods to improve reproducibility and validation of predictive models
  • Re-envisioning how the electronic health record is presented to physicians and patients
  • Applying machine learning to inform population health management

Research Team

Shiwen Shen

Shiwen Shen

PhD Candidate

Profile
Nicholas Matiasz

Nicholas Matiasz

PhD Candidate

Profile
Edgar Rios

Edgar Rios

PhD Candidate

Profile
Nova Smedley

Nova Smedley

PhD Candidate

Profile
Panayiotis Petousis

Panayiotis Petousis

PhD Candidate

Profile
Name  Department, University
Aditya Srinivasan Computer Science, UCLA Undergraduate
Albert Chern Computer Science, UCLA
Allen Wu Computer Science, UCLA
Andy Reyes Computer Science, UCLA
Curtis Crowther Biology, FAMU UC-HBCU, Undergraduate
Ebrain Mirambeau Mathematics, UCLA
Ellen Peterson Biology, FAMU UC-HBCU, Undergraduate
Evan Zhen Computer Science, UCLA
Gene Auyeung Computer Science, Rice CDSC
Jose Peres Computer Science, UCLA
Robert Seniors Biology, FAMU UC-HBCU, Undergraduate
Jiayu Sun Peking University
Roxanne Loo Bioengineering, UCLA Associate Bioengineer, Medtronic
Uche Ononuju Biology, FAMU UC-HBCU, Medical Student, Wayne State University
Yousef Mohammad Computer Science, UCLA Quality assurance analyst, ITG Inc.
Yunming Zhang Computer Science, Rice CDSC
Juan Anna Wu, PhD

Juan Anna Wu, PhD

Academic Program Advisor, Cedars-Sinai, CA

Profile
Kyle Singleton, PhD

Kyle Singleton, PhD

Research Associate, Mayo Clinic, AZ

Profile

Research Projects

  • Adapting and validating novel decision models and uncertainty analysis to discover optimal strategies for individual patients

    Algorithms for discovering optimal strategies and novel linkages from heterogeneous clinical data

    Physicians routinely face the challenge of reasoning on complex, multimodal data under uncertainty. Algorithms are needed to discover: what information is relevant to make a decision; what test/treatment is most effective for a given individual; and how to present inherent uncertainty in model predictions to the physician or patient.

    Related Projects

    • Using sequential decision making approaches to discover optimal strategies for breast cancer screening
    • A Bayesian approach for continuous Markov models: Towards individualized lung cancer screening recommendations
  • Formalizing literature for treatment selection and experiment planning

    Creation of a logical representation and probabilistic approach to summarize and match evidence to patients

    In 2011, the Institute of Medicine released a report on precision medicine, revealing that a “knowledge network” is needed to integrate the results of basic research with data being collected at the point-of-care (National Academies Press, 2011). In this light, my research addresses two intertwined issues: the lack of a formal representation to capture experimental results described in the literature, particularly the context in which the experiment was conducted; and the difficulty of linking relevant evidence reported in the literature to individual characteristics of patients. I developed a formal model to represent numerical evidence (e.g., statistical analyses, measured outcomes) reported in the clinical trial literature in a machine interpretable manner, facilitating the automated quality assessment of studies and the matching of patients to relevant treatments (Hsu et al, 2014; Tong et al, 2014; Tong et al, 2016).

    Related Projects

    • Representing and synthesizing results from causal experiments for hypothesis generation and experiment planning
    • Methods for structuring and visualizing the results of clinical trial papers
  • Improving the generalizability of predictive models

    Metrics enabling the assessment and reuse of published predictive models in different contexts

    While papers describing predictive models for clinical decision support have been widely published, usage of these models in practice has been poor. Most model evaluations focus on the internal validity of a model (i.e., evaluation on a single sample in a single setting), but such evaluations are insufficient to predict how well the model would perform for observations collected at different institutions (Singleton et al, 2012).

    Related Projects

    • A formalized approach to evaluate the transportability of a predictive model using simulation methods
    • PREMIERE: A platform for sharing and evaluating probabilistic models
  • Re-envisioning the electronic health record

    Context-sensitive presentation of electronic health records based on the user's information needs

    Routine patient care often results in a variety of data (e.g., text, imaging, numeric values) being captured as part of the patient’s medical record. Poor integration between the various data sources (e.g., labs, imaging, pathology) and the inclusion of excess information can hamper physicians from noticing important findings. My work in clinical data visualization has been focused on utilizing domain knowledge encoded in graph-based structures (e.g., biomedical ontologies, graphical models) to adaptively filter and tailor the presentation of complex, longitudinal health data to target users (e.g., physicians versus patients).

    Related Projects

    • AdaptEHR: An adaptive approach to presenting problem-centric electronic health record data
  • Data mining for population health management

    Learning from electronic medical records to improve the accuracy and efficiency of delivering care through imaging

    The shift from fee-for-service to value-based reimbursement underscores the need to objectively assess the value of radiologic interpretations, including the accuracy with which a patient’s condition is diagnosed and the time saved to reach the appropriate diagnosis. Population health management attempts to leverage the large amounts of clinical data being routinely collected during patient care to answer questions related to utilization, efficiency, accuracy, and value of diagnostic procedures. My efforts are focused on demonstrating the role and value of radiology exams.

    Related Projects

    • RadACC: Improving the diagnostic accuracy of radiology by facilitating a data-driven assessment of radiologist-provided impressions with downstream “precision” diagnoses
    • Assessing the cost of radiology using time-driven activity based costing measures