Incubating Digital Healthcare Startups
The Center for Machine Learning and Health (CMLH) at Carnegie Mellon University is one of two centers launched under the umbrella of the Pittsburgh Health Data Alliance, formed in 2015 to unite Carnegie Mellon’s unrivaled applied-computing capabilities, the University of Pittsburgh’s world-class health sciences research, and UPMC’s clinical care and commercialization expertise.
Our Focus Areas
The CMLH supports great science and engineering that can lead to innovative health solutions and new businesses. The Center's three focus areas are:
Connect and coordinate the health system to empower clinicians to provide high-quality care in any setting.
Develop solutions that allow consumers to access medical services and information anytime, anywhere, and to engage in all steps of the healthcare journey.
Infrastructure and Efficiencies
Enhance resource allocation, service levels, and care pathways to coordinate and manage the cost of care.
2019 Funding Opportunities
From bench to bedside: the CMLH funds projects that strive to bridge the gap between research and practice. All funded work at CMLH will have a clear line of sight to commercial application. Although many CMLH projects will involve data analytics and machine learning, our approach is technology agnostic. We welcome proposals that involve human-computer interaction, language technologies, information systems, computer graphics, computer vision, artificial intelligence, robotics, electrical engineering, economics, psychology, sociology, public policy, business administration, law, design, and any other disciplines that apply to healthcare.
CMLH initially provides research projects with approximately one year of funding. After one year, projects may attract more funding to refine the technology and/or its development for commercialization.
In addition to offering funding, CMLH provides support for funded projects in the form of datasets, access to patients and doctors for empirical validation of new concepts, and entrepreneurial mentorship.
Awards are intended to support research that transforms healthcare from transactional and experience-centered to data-driven. Once selected for an award, the CMLH and UPMC Enterprises will help identify potential clinical and data-transaction partners and provide guidance related to commercialization activities, as needed.
Proposal submissions are open only to Carnegie Mellon faculty in all units of the university. All proposals should be submitted via email to: firstname.lastname@example.org.
Download an overview of our Center for Machine Learning and Health projects here:
The last decade of research has established that Cortical Spreading Depolarizations (CSDs), or “Brain Tsunamis,” play an important role in many disorders, including Traumatic Brain Injury (TBI), stroke, hemorrhage, and migraine, that collectively affect more than a billion people worldwide and are major causes of death and disability. CSDs are waves of neurochemical changes that propagate across the brain surface and are thought to mediate secondary brain damage following an injury or event.
The only reliable way to measure CSDs now is an invasive method using electrodes; it is costly, risky and inappropriate for patients with certain contraindications. In addition, detection is slow and requires a physician to visually inspect the data. Diagnosis and treatment in real-time thus is not realistic in most clinical settings.
Our solution is a Noninvasive Platform for Automated CSD Detection and Suppression. Our multidisciplinary team will develop algorithms and techniques to non-invasively suppress CSDs by using online detection and ensuing stimulation. This is the critical first step for discovery of novel therapeutic strategies and would eliminate the need for invasive monitoring. Real-time detection would allow rapidly iterated, individualized care, timely intervention, and enhanced patient outcomes.
This project is led by Pulkit Grover, PhD, Assistant Professor of Electrical and Computer Engineering at Carnegie Mellon University; Shawn Kelly, PhD, Research Faculty at Carnegie Mellon University; Marlene Behrmann, Professor of Psychology at Carnegie Mellon University; Michael Tarr, PhD, Department Head of Psychology at Carnegie Mellon University; Jonathan Elmer, MD, Assistant Professor of Critical Care Medicine and Emergency Medicine at University of Pittsburgh; and Lori A. Shutter, MD, Vice Chair of Education at University of Pittsburgh.
The Clinical Genomics Modeling Platform is an engine for easily building precision-medicine models for various diseases and populations. Triage algorithms, for instance, might help to determine if patients with a certain disease should be sent home with monitoring or sent to the intensive care unit.
Their project focuses on the understanding of the relationship between genetic variation and medical outcomes in a large population, which is key to realizing the vision of personalized medicine. Efforts are now underway to obtain the genome sequences of thousands of individuals, and as the cost of sequencing continues to drop, it will become routine to sequence patients in a medical setting. However, a number of computational and practical challenges remain in the way of using genomic sequencing for clinical decision making.
In this project, Dr. Carl Kingsford and Dr. Christopher Langmead are aiming to increase the usage of predictive systems based on machine learning techniques and genomics through the development and commercialization of a computational system that will
- Make predictive models easy to build for clinical researchers
- Make predictive models easy to share (sell) and apply
- Make results of models easy to understand
This project is led by Carl Kingsford, PhD, and Christopher Langmead, PhD, both work within Carnegie Mellon University’s Computational Biology Department.
Readmission after complex cancer surgery is common, with studies reporting between 15% and 50% of patients being readmitted within 30 days of discharge from the hospital. Readmissions are associated with increased healthcare costs, poor clinical outcomes including increased risk of infection and early mortality, and patient and family stress and suffering. Prior research has identified demographic and clinical predictors of readmission, but the role of patient-centered behavioral processes remains relatively unexplored. Identifying behavioral factors associated with readmission, which might include physical activity levels, sleep habits, and social contacts, could highlight ways to prevent readmissions and empower patients to take a more active role in their recovery.
The project team is collaborating to take a generalizable and scalable approach that holistically looks at patients’ behavior before and after surgery and identifies routines in this behavior. These routines will form the basis of their understanding of the predictive factors for readmissions, particularly while patients are still in the hospital after surgery or other treatments.
Harnessing technology to monitor these and other behavioral risks before surgery, during inpatient recovery, and during the critical transition from hospital to home will advance the field in a number of ways:
- Reliably assessing and computationally modeling behavioral risks associated with readmissions to improve risk stratification
- Identifying optimal targets and timing for behavioral intervention to reduce preventable readmissions
- Providing data that may be helpful and motivating for patients and that could inform clinical decision making to improve quality of care
This project is led by Anind Dey, PhD, Charles M. Geschke, PhD, Director of the Human-Computer Interaction Institute at Carnegie Mellon University, and Carissa Low, PhD, Assistant Professor of Medicine and Psychology, Biobehavioral Oncology Program, University of Pittsburgh Cancer Institute.
A collaborative team from CMU, Pitt, and UPMC is developing image analysis software to examine slide images of placentas following delivery and identify abnormalities that require review by a pathologist. Abnormalities found in these images can provide critical information about the health of the infant and mother.
Though the placenta receives a routine examination in the delivery room, experienced pathologists – especially those with perinatal subspecialty expertise – are not always available to handle the volume of incoming placentas requiring examination. The software being developed in this project thus would reduce the risk that physicians would miss detecting otherwise avoidable complications.
The researchers will apply deep learning techniques to create an algorithm that will support high-throughput examination of whole-slide images of placentas and identification of placentas that most need a pathologist’s attention. This approach will be enabled by using the digitized whole slide imaging dataset of placenta slides at UPMC, together with the current rich dataset available from the Magee Obstetric Maternal & Infant (MOMI) Biobank of placenta images coupled with pathology metadata, clinical data regarding the delivery, and long-term data about the infant’s health.
This project is led by Jonathan Cagan, PhD, Professor of Mechanical Engineering (and Interim Dean of the College of Engineering) at Carnegie Mellon University, Philip R. LeDuc, PhD, Professor of Mechanical Engineering, Carnegie Mellon University, Janet M. Catov, PhD, MS, Associate Professor, Department of Obstetrics, Gynecology and Reproductive Sciences and the Department of Epidemiology, University of Pittsburgh, and Liron Pantanowitz, MD, Vice Chairman for Pathology Informatics at UPMC and Director of Cytopathology at UPMC Shadyside.
A non-invasive device that analyzes sounds from the intestinal tract could become a powerful new tool to help physicians diagnose and monitor a variety of gastric illnesses, such as acute pancreatitis, bowel obstructions, irritable bowel syndrome, inflammatory bowel disease, and Crohn’s disease.
Carnegie Mellon researchers, collaborating with UPMC gastroenterologists, will be developing, testing, and performing clinical research on a wearable sensor array to detect intestinal sounds – a new vital sign for the “gut.” Its impact could be substantial: In the U.S. alone, digestive diseases afflict 60-70 million people annually with treatment costs totaling more than $100 billion.
Using machine learning techniques, the researchers will develop a means of interpreting the acoustic signals to examine the determinants of gut activity and activity suppression, the ability to identify gastric disorders on the basis of sounds, and the device’s therapeutic potential for enabling gut regulation via biofeedback. The project will also develop classification procedures that use these features to distinguish between the various types of gastrointestinal activity.
The research team includes George Loewenstein, PhD, Professor of Economics and Psychology, Department of Social and Decision Sciences and Ali Momeni, PhD, Associate Professor of Art, College of Fine Arts, along with Rich Stern, PhD, Professor Electrical & Computer Engineering, Max Gsell, PhD, Associate Professor, Department of Statistics and Valerie Ventura, PhD, Associate Professor, Department of Statistics.
Medical diagnostic errors impact 12 million adults each year in the US. The number of deaths due to medical diagnostic errors is the third leading cause of death, equivalent to the crash of a large aircraft every day, based on estimates in the US alone. A key reason why diagnostic errors are made – even by the best clinicians in highly reliable organizations – is the increasing complexity of the diagnostic process, with over 10,000 diseases and 5,000 laboratory tests to choose from.
This project focuses specifically on preventing coding and billing errors. To address this cognitively complex problem, the team is developing an engine that will predict likely diagnosis codes based on information available in a patient’s electronic health record. Specifically, the solution will review both structured and unstructured data, such as clinical notes, and apply a machine learning-based mapping from these data to specific diagnosis codes.
When the diagnosis code suggestions are acknowledged and accepted, these can then also flow into billing codes through electronic medical records or other systems. The solution will thus aid clinicians in their medical management, decision support, accurate documentation, billing and quality improvement.
This project is led by Pradeep Ravikumar, PhD, Associate Professor of Machine Learning at Carnegie Mellon University, and Jeremy Weiss, MD, who works within the Heinz College at Carnegie Mellon University and the Department of Biomedical Informatics at University of Pittsburgh.
This project is determining if coordinated care decisions could be improved by analyzing and monitoring how patients move through the healthcare system.
They will be focusing on healthcare trails – time-ordered sequences in which a patient obtains services such as dialysis, blood work, or psychological counseling. These trails will be analyzed to identify those that are commonly used by patients with similar conditions and those that most often improve patient conditions. Factors such as a patient’s age, healthcare coverage, and medical condition can affect the flow of these services, and differences in flow can be associated with different patient outcomes.
The goal is to develop, test, and assess algorithms and interfaces for analyzing hospital admissions data and for providing feedback to doctors and caregivers through automated patient tracking and notification systems. By analyzing the timing of admits, discharges, and transfers, the application could help inform care givers about how a patient’s care is being managed and detect when a patient might be at risk because of the healthcare trail they are on.
This team is led by Kathleen M. Carley, PhD, of the Institute for Software Research and L. Richard Carley, PhD, of the Electrical and Computer Engineering Department.
A team from Carnegie Mellon, the University of Pittsburgh, and UPMC is working on a project to evaluate the ability of cameras, inertial measurement units, and other sensors, in combination with machine learning (ML) algorithms, to assess patients’ movement therapy exercises in the home. The long-term goal is to increase the quality and efficacy of physical therapy by providing patients with automated, in-home monitoring, near real-time feedback on exercise performance, and feedback to providers when issues arise outside of the PT setting. This will enable providers to prioritize clinic time for patients whose recovery has stalled while avoiding unnecessary appointments for those who are progressing satisfactorily.
Leading the team’s machine learning evaluations from Carnegie Mellon’s School of Computer Science are Jessica Hodgins, PhD; Dan Siewiorek, PhD; Asim Smailagic, PhD; and Andrew Whitford, PhD. Keelan Enseki from the UPMC Center for Rehab Services and Adam Popchak, PhD, from the University of Pittsburgh’s Department of Physical Therapy, are leading the clinical stages of the project.
The costs of preterm births to the healthcare system and society are exceedingly large, and the incidence of preterm births is on the rise despite medical advances. This has created a need to provide reliable, personalized information to pregnant women about how they can reduce their risk of premature births. In response, researchers have created the “MyHealthyPregnancy” smartphone app.
The app applies statistical machine learning techniques to comprehensive pregnancy data sets to improve the app’s patient-specific risk predictions of adverse pregnancy outcomes. It also employs decision science techniques to extend the app to provide post-partum education and behavioral feedback to minimize infant mortality outcomes. The team, which has performed pilot studies of the concept, expects this work will result in a product that can be tailored to both the UPMC health system and developed into an off-the-shelf application for the general population of pregnant women.
The project is led by Hyagriv Simhan, MD, of Magee-Women’s Hospital and Tamar Krishnamurti, PhD, and Alexander Davis, PhD, of Carnegie Mellon University.
The healthy brain maintains a relatively constant blood flow to the brain even when there are changes in blood pressure or intracranial pressure (ICP). The mechanism preserving blood flow is called cerebral autoregulation (CA), which is known to be impaired in a variety of diseases, such as diabetes, Parkinson’s disease, stroke, and traumatic brain injury (TBI).
ICP now can only be measured by placing an invasive pressure sensor inside the brain.
This project seeks to establish an optical technique called near-infrared spectroscopy (NIRS) as a non-invasive method to monitor ICP in humans. Development of a non-invasive ICP sensor would optimize patient treatment in cases where invasive ICP is not possible or could be dangerous to the patient.
This project team consists of Jana Kainerstorfer, PhD, Assistant Professor of Biomedical Engineering at Carnegie Mellon University and Elizabeth Tyler-Kabara, MD, Neurological Surgery at UPMC Children’s Hospital of Pittsburgh.
Despite screening technologies and public health efforts that have improved the detection of early-stage cancers, methods to reliably predict which of these cancers will progress to more aggressive stages, metastasis, and ultimately patient mortality are lacking.
The project team is working on a suite of software tools that clinicians will use to improve cancer diagnosis and therapeutics based on the molecular signatures of the patient’s tumor genome.
The goal of the project is to develop novel models, algorithms, and software tools to better understand the origin and evolution of tumor cells and how a patient’s tumors are likely to progress. This information would contribute to personalized, precision treatment of cancer.
The project is led by Russell Schwartz, PhD, Professor of Biological Sciences and Computational Biology and Jian Ma, PhD, Associate Professor of Computational Biology, both with the School of Computer Science.
As health data is increasingly digitized, the need to protect patient privacy is unprecedented. The Programming Framework for Managing Patient Privacy team is collaborating on an approach to protect patients’ health records and computations involving sensitive patient data from being leaked in a breach.
Project researchers propose a programming model and software framework for managing the privacy of health data. The programming model allows programs to attach privacy policies directly to the data, rather than requiring the programmer to implement the correct privacy checks across code that uses the data. Attaching privacy policies to the data enables programs to be policy-agnostic and to be updated independently of the specification and implementation of privacy policies. The system tracks the flow of data, as well as the policies, so the programmer need not do so. This approach facilitates auditing, as policies may now be centralized and to be implemented only once, instead of as checks across the program. Auditors may also leverage the fact that the framework, rather than the programmer, is now responsible for correctly implementing the checks.
This project is the first step in a long-term plan to improve patient care and promote scientific discovery by allowing sensitive health data to be shared safely. This plan involves 1) applying prior work on policy- agnostic programming for information flow policies in the context of patient health records, 2) extending and combining prior work to support more expressive policies, and 3) building a practical framework that can be used to safely share sensitive data.
The project team consists of Jean Yang, PhD, Matt Fredrikson, PhD, and Jian Ma, PhD, from Carnegie Mellon’s School of Computer Science.
Over 150,000 Americans are treated in hospitals each year after cardiac arrest, virtually all of whom are initially comatose. Once their hearts are revived, most such patients eventually die in the hospital from brain injury, but a sizable minority will awaken and have a good neurological recovery. With current technology, no sign, symptom, or test result in the first 72 hours after cardiac arrest is sufficient to preclude a favorable recovery. Prognosis often remains uncertain for days or even weeks thereafter. The project team is developing technology that will use advanced signal processing and modeling methods to allow accurate neurological prognostication sooner than currently possible.
Project team members are Daniel Nagin, PhD, at Carnegie Mellon University’s Heinz College, and physicians Jonathan Elmer, PhD, and Cliff Callaway, MD, from the University of Pittsburgh Medical School.
Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection that has a high mortality rate and is a large burden on the health care industry, constituting over $24 billion annually.
This team will use machine learning and artificial intelligence methods to develop analytic tools to identify sepsis earlier and define subgroups of patients who are at high risk for sepsis and share other traits. Successful completion of this research will advance efforts to identify subgroups with sepsis for whom particular treatments are more effective, thereby reducing morbidity and mortality. It will also support the development of physician-competitive, health records-based risk scores that can be used for risk stratification and for early warning clinical decision support. This proposal leverages an established collaboration between experts in machine learning and sepsis at CMU, Pitt, and UPMC.
This project is led by Jeremy Weiss, PhD, who works within the Heinz College and Machine Learning at Carnegie Mellon University and the Department of Biomedical Informatics at University of Pittsburgh; and Christopher Seymour, MD, who works within the Critical Care and Emergency Medicine at University of Pittsburgh.
Depression is a leading cause of disability worldwide. Effective, evidence-based treatments for depression exist but many individuals suffering from depression go undetected and therefore untreated. Efforts to increase the accuracy, efficiency, and adoption of depression screening thus have the potential to minimize human suffering and even save lives.
Recent advances in computer sensing technologies provide exciting new opportunities to improve depression screening, especially in terms of their objectivity, scalability, and accessibility. Professor Morency and Dr. Szigethy are collaborating to develop sensing technologies to automatically measure subtle changes in individuals’ behavior that are related to affective, cognitive, and psychosocial functioning. Their goal is to develop and refine computational tools that automatically measure depression-related behavioral biomarkers and to evaluate the clinical utility of these measurements.
This project is led by Louis-Philippe Morency, PhD, Finmeccanica Associate Professor of Language Technologies at Carnegie Mellon University and Eva Szigethy, MD, Professor of Psychiatry and Director of Behavioral Health with the Chief Medical and Scientific Office at University of Pittsburgh.
Joe Marks, PhD
Joe Marks, PhD, Executive Director of CMLH, leads the strategic initiatives, business development, and negotiations for the Center. Joe brings …
Joe Marks, PhD, Executive Director of CMLH, leads the strategic initiatives, business development, and negotiations for the Center. Joe brings a diverse background to CMLH that includes leadership roles at Disney Research and Mitsubishi Electric Research Labs (MERL), and involvement in multiple start-ups.
He earned his undergraduate and PhD degrees from Harvard University. His publications have focused mostly on topic areas that include computer graphics, human-computer interaction, and artificial intelligence, but he’s happy to work in new areas as the opportunity arises.
Carl Kingsford, PhD
Chief Science Officer
Carl Kingsford, PhD, is a Professor in the Computational Biology Department in the School of Computer Science at Carnegie Mellon …
Carl Kingsford, PhD, is a Professor in the Computational Biology Department in the School of Computer Science at Carnegie Mellon University. His research group develops efficient algorithms for analysis of large-scale biological data, particularly high-throughput genomic sequencing experiments. He and his group are the authors of several widely used, open-source software packages for genomic analysis. More information about his research can be found at http://kingsfordlab.cbd.cmu.edu. He is the recipient of an NSF CAREER award, a Sloan Research Fellowship in Computational and Molecular Biology, and he is one of 14 Data-Driven Discovery Investigators funded by the Gordon and Betty Moore Foundation selected from across all areas of science. He received his PhD in Computer Science from Princeton University.
Marketing and Commercialization Advisor for the CMLH
Distinguished Service Professor of Digital Media and Marketing, Heinz College, Carnegie Mellon University