Assessing Mammography AI

While routine screening has been shown to decrease breast cancer mortality in multiple randomized controlled trials, mammography is limited by subjective human interpretation. Recent advances in improved computer processing power, cloud-based data storage capabilities, and availability of large imaging datasets have led to renewed excitement for applying artificial intelligence (AI) to mammography interpretation. A collaboration between the University of Washington (Christoph Lee) and UCLA (Joann Elmore, Arash Naeim, William Hsu), this project is undertaking a unique academic-industry partnership to validate, refine, scale, and clinically translate our proven 2D mammography AI algorithm to 3D mammography interpretation. We are validating an existing algorithm for 2D mammography using UCLA’s Athena Breast Health Network. We are enhancing this 2D AI algorithm with expert radiologist supervision and will be examining the impact of adding novel non-imaging data parameters, including genetic mutation and tumor molecular subtype data, to help train the AI model to identify more clinically significant cancers.

We are using several novel technical algorithmic approaches to scale from 2D to 3D mammography that in our preliminary studies have shown improved accuracy beyond radiologist interpretation alone. A series of interpretive studies are planned to identify the optimal interface between “black box” outputs and radiologist interpreters, which remains an understudied topic. Our new end-user tool aims to help tip the balance of routine screening towards greater benefits than harm.

Advancing mHealth Informatics

The Los Angeles (LA) PRISMS Center aims to be the leader in the development and application of mobile health (mHealth) technologies that deepen our scientific understanding and clinical management of diseases. Bringing together leading experts from UCLA and the University of Southern California (USC) in biomedical informatics, computer science, wireless health, environmental health science, and pediatrics, this Center supports innovative end-to-end software infrastructure for sensor-based health monitoring.

Focusing on pediatric asthma, our Center’s vision and research is motivated by the following question: what if you could predict ahead of time, for a specific asthma patient, the potential for exacerbation and thus mitigate – if not prevent – the event? Any system with this ability must integrate the growing array of available physiologic and environmental data from sensors, and place such data into context to elucidate the patient’s state and specific situation. The system must be able to act sufficiently quickly on sensed data to make timely recommendations, and end user compliance with system usage must be high to effect change. Our solution, the <strong>Biomedical REAl-Time Health Evaluation for Pediatric Asthma (BREATHE)</strong> platform, provides an extensible framework for the deployment of data collection protocols; secure data collection from sensors to a mobile device; integration of additional contextual information; and real-time analysis. Importantly, usability is a central consideration in the design of BREATHE, reflected in an iterative design/evaluate/refine process. To build and assess BREATHE, the Center comprises three closely coordinated efforts: Project 1 – Integrated Sensing from the Device to the Cloud, which establishes APIs for automatically gathering information from a device and local sensors, communicating with commercial and PRISMS U01 sensors and U24 coordinating data center; Project 2 – Integrating &amp; Visualizing Clinical, Environmental, and Sensor Data, which focuses on combining data acquired from the U24 data center with contextual information (e.g., regional air quality, clinical elements from the patient’s electronic health record, etc.) with real-time processing and analysis infrastructure; and Project 3 – Real-time Asthma and Air Pollution Project (Asthma APP), which develops a framework for evaluating system performance and real-world field testing of the platform for self-management and early interventions. Collectively, these Projects’ efforts realize BREATHE, changing how we interact with pediatric asthma patients and their caregivers to actuate a better understanding of the disease and improve adherence, and to achieve more personalized medicine through more detailed, objective measurements of an individual’s daily activities and surroundings.

Data-driven Diagnostic Decision Support

We are implementing a prototype next-generation decision support system called SmartDx, which learns the optimal sequence of diagnostic tests tailored to a patient’s unique characteristics and circumstances. The goal of this project is to optimize clinical pathways based on diagnostic accuracy, timeliness, and cost for individual patients using routine data collected longitudinally in the electronic health record. We are developing and validating machine learning algorithms that adapt by discovering the relevant clinical features that are most informative in identifying the appropriate diagnostic test for the individual. The tool personalizes the sequences of tests based on the availability of new information, optimizing based on the cost, timeliness, and accuracy of test results. Our efforts are driven by practical decision support questions related to optimal paradigms for breast and lung cancer screening, which are at the forefront of radiology today.

The project is organized around three key tasks: 1) development of novel adaptive learning methods to discover the most informative features that are predictive of subsequent actions taken in real-time; 2) exploration of deep reinforcement learning approaches that not only discovers what is the next best diagnostic test to order but also identifies additional information that is needed to make a definitive diagnosis; and 3) assessment of methods for making decision support models more transparent by communicating the rationale and uncertainty associated with model predictions. Our objectives are to: 1) determine what combination of diagnostic procedures (e.g., imaging, labs, biopsy) should be used to achieve an accurate and timely diagnosis – and in what sequence; and 2) demonstrate that learning such pathways can be done using real-world clinical data, allowing our methodology to be applied in realistic scenarios that require learning from incomplete and inconsistent data.

Improving Prostate Cancer Diagnosis

Prostate cancer is the second leading cause of cancer death in American men, accounting for 26% of new cancer diagnoses and 9% of cancer deaths in men. Active surveillance, radical prostatectomy and radiotherapy are commonly used treatments for clinically localized prostate cancer. However, current risk stratification methods cannot be used effectively to avoid subjecting patients with clinically indolent cancers to unnecessary interventions, causing significant morbidity and cost.

The research objective of this R21 is to develop novel techniques using multiparametric magnetic resonance imaging (mp-MRI) and MRI-ultrasound (US) fusion guided biopsy data that provide discriminatory power in distinguishing indolent versus clinically significant prostatic adenocarcinoma based on non-invasive imaging. We are implementing a multi-instance learning (MIL) based convolutional neural network (CNN) model for clinical prostate mp-MRI sequences to generate new quantitative imaging features representative of the underlying tissue. Our MIL-CNN model accomodates ground truth labels from pathology whole mount specimens, as well as MRI-US fusion biopsy results. Hierarchical CNN features will be used to predict voxel-level cancer suspicion, thereby enabling a novel method for performing “imaging biopsy.” Finally, voxel-level suspicion maps will be aggregated into patient-level quantitative imaging biomarkers and combined with clinical data to create a multimodal nomogram for performing risk stratification.

CDS for Indeterminate Pulmonary Nodules

The rollout of low-dose computed tomography (LDCT) lung screening programs is accelerating in the United States, aiming for earlier detection of lung cancer to improve long-term survival. However, a consequence of such imaging programs is the increased discovery of indeterminate pulmonary nodules (IPNs). Significant questions remain around the effective management of screen- and incidentally-detected IPNs: while many are benign, a fraction will go on to become cancerous.

The objective of this R01 is the development of a clinical decision support tool for the management of screen- and incidentally-detected IPNs. We address two key challenges: 1) the development of a continuous-time model for predicting how the IPN will evolve; and 2) the use of this prediction to determine a series of actions over time that will optimize (screening) outcomes for the individual. We first explore the development of a continuous time belief network (CTBN), a temporal probabilistic model to predict the likelihood of a patient to develop lung cancer. The probabilities computed through the CTBN are subsequently input into a partially-observable Markov decision process (POMDP) to guide IPN management decisions.

Making Biomedical ML Reproducible

The confluence of machine learning (ML) data-driven approaches and increased computational power, alongside access to the wealth of electronic health records (EHRs) and other emergent types of data (e.g., omics, imaging, mHealth), are accelerating the development of biomedical predictive models. Such models range from traditional statistical approaches (e.g., regression) through to more advanced deep learning techniques (e.g., convolutional neural networks, CNNs), and span different tasks (e.g., biomarker/pathway discovery, diagnostic, prognostic, etc.). Two issues have become evident: 1) as there are no comprehensive standards to support the dissemination of these models, scientific reproducibility (vs. replicability) is problematic, given challenges in interpretation and implementation; and 2) as new models are put forth, methods to assess differences in performance, as well as insights into external validity (i.e., transportability), are necessary. Tools moving beyond data sharing and model “executables” are needed, capturing the information needed to fully reproduce a model and its evaluation. The objective of this R01 is the development of <strong>PREMIERE (PREdictive Model Index and Exchange REpository)</strong>, an informatics standard supporting the requisite information for scientific reproducibility for statistical and ML-based biomedical predictive models.

Understanding Chronic Kidney Disease

Significant health disparities exist in chronic kidney disease (CKD), CKD progression, and end stage renal disease (ESRD) in ethnically diverse populations. African Americans (AAs) have approximately 25% higher prevalence of CKD, and 3x higher rates of ESRD compared to their counterparts. Moreover, a portion of these individuals exhibit rapid decline, progressing much faster than other comparable patients. Using a unique repository build from the electronic health records of &gt;10 million individuals seen at UCLA and Providence Health System, we are building new predictive models to better understand CKD progression and issues related to AAs. Using machine learning (ML)-based methods, we are developing CKD and eGFR trajectory models over time, stratified across age, gender, and different racial/ethnic groups. These models will provide insight into the factors that influence CKD, resulting in worsening outcomes. To ensure that these models translate into effective interventions, we are also conducting focus groups with primary care physicians to elicit their perspectives on existing and designed models to reduce CKD health disparities.