Projects

Welcome to the Clinical AI Research Lab! Explore our AI-driven healthcare projects, from wound assessment and heart disease detection to gait analysis, infectious disease monitoring, neurolinguistics, and pain management—advancing healthcare with innovative AI solutions.

Smartphone Wound Assessment and Decision Support (SmartWAnDS)

  • Chronic wounds affect 10.5 million US patients with prevalence in the elderly, take up to 1 year to heal and cost $28 billion annually. Currently, up to 80% of wound patients are treated by nurses who visit their homes periodically (~weekly) to assess and document the wound’s healing progress, change dressings, apply medicines, and refer problematic cases to experts. However, a shortage of visiting nurses with wound expertise is resulting in in-home wound care issues including: 1) Wound type identification errors:  2) Pressure Ulcer (PU) stage identification errors: in 71% of wound cases. 3) Manual wound measurement errors, overestimating wound size by up to 44% 4) Wound healing progress assessment: that has high inter- and intra- rater variability, and 5) Late referrals to experts: resulting in avoidable limb amputations, a surgery that costs $33,499. In fact, 85% of limb amputations are preceded by a wound. Care errors often lead to nurse malpractice lawsuits with over 17,000 on PUs alone with a mean payout of $4 million by nursing homes. Amputations are also costly with a lifetime cost of $509,272 per amputation. Moreover, post-amputations, patients die within 5 years.
  • Our team has researched and developed SmartWAnDS, a smartphone-based wound image analyses and decision support system that utilizes AI to address all five in-home wound care issues listed above. SmartWAnDS analyzes a smartphone wound image taken by a visiting nurse to 1) Identify the wound type, 2) Identify the PU stage 3) Measure the wound size 4) Score the wound’s healing progress by analyzing wound size, depth, amount and type of granulation tissue, amount and type of necrotic tissue, wound edges and periulcer skin viability, 8 attributes defined by the Photographic Wound Assessment Tool (PWAT), a validated wound healing scoring rubric, and 5) Recommend whether the wound patient should be referred to experts in the clinic. Competing wound systems focus on estimating low level parameters such as wound size from an image, to reduce measurement errors and speed up documentation that is currently tedious and time-consuming. However, interpreting low level wound parameters and making care decisions still requires expertise, a challenge faced by non-expert nurses in patients’ homes. SmartWAnDS is the first to use AI to generate the validated PWAT healing score and recommending when patients should be referred to experts and best evidence-based care.

Cardiovascular Disease Assessment

Cardiovascular Diseases (CVDs) are the number one killer of humans globally. Several issues can be addressed using AI including improving the consistency of experts, reducing cost and improving access to minority groups. Example CVD assessment projects include:

  • Video Action Recognition Deep Learning Models to Detect Hypertrophic Cardiomyopathy (HCM) in echocardiogram videos (heart ultrasounds): Hypertrophic cardiomyopathy (HCM) is the most common inherited cardiovascular disease (CVD) but is significantly underdiagnosed with only 13% of cases identified [8]. HCM causes an abnormal thickening of the heart muscle, which can result in severe adverse outcomes including Atrial Fibrillation (AF), Heart Failure (HF) and Sudden Cardiac Death (SCD) [17]. Accurate, timely HCM diagnosis is needed to identify candidates for lifesaving therapies such as recently FDA-approved (2022) Mavacamten [167,168], which improves symptoms, slows HCM progression, and can decrease the need for surgery in HCM patients with Left Ventricular Outflow Tract (LVOT) obstruction.
  • Video Action Recognition Deep Learning Models to Classify Cardiac Ejection Fraction (EF): The EF of a heart is the ratio of the amount of blood pumped out of the heart-ventricle to the volume of blood entering the ventricle at the start of each heartbeat. EF is an important parameter for assessing cardiac function. An echocardiogram, an ultrasound of the heart that generates videos of the beating heart, is widely used to assess cardiac function and diagnose CVDs.

Smartphone Gait Analyses

Gait, or the way a person walks, is a reliable bio-measure of many conditions including intoxication by alcohol, impairment by marijuana and parkinsons disease. We have researched and developed various machine learning models that analyze data from sensors such as accelerometers and gyroscopes that are built into many modern smartphones and smarwatches, to perform gait analyses.

  • AlcoGait/WeedGait: Driving is a dynamic activity, which requires quick reflexes and decision making in order to respond to sudden changes in traffic conditions. Alcohol or/and marijuana consumption impairs motor and cognitive skills, and causes many driving-related accidents annually. Passive methods of proactively detecting drivers who are too drunk to drive in order to notify them and prevent accidents, have recently been proposed. The effects of alcohol on a drinker’s gait (walk) is a reliable indicator of their intoxication level.
  • Parkinsons Detection from Gait: Nearly 1 million people in the USA and 10 million people worldwide are living with Parkinson’s disease (PD). The progression of the disease can be inferred from the changes in the patients’ gait to inform early intervention.

DARPA Warfighter Analytics for Smartphone Healthcare

Warfighters face an increased exposure to various ailments such as Traumatic Brain Injury (TBI) and infectious diseases. Digital biomarkers are smartphone-sensable user behaviors that can reliably indicate the health status, ailment symptoms and condition of the smartphone user. For example, an ailing smartphone user may exhibit reduced daily step counts or stay longer in sedentary activity states during their day.

Funded by the DARPA Warfighter Analytics using Smartphones for Health (WASH) project, our team is researching and developing machine/deep learning algorithms that synthesize reliable smartphone biomarkers that enable continuous, real-time assessment of TBI and infectious diseases afflicting warfighters by leveraging data unobtrusively captured from smartphone sensors. We are proposing deep learning models that detect the user’s current situation, smartphone biomarkers, and also generate a bioscore that we define as the likelihood that a given user has a specific condition (e.g. TBI or an infection disease).

Infectious Disease Detection from Wearable Data:

Covid-19, a recently discovered Influenza Like Illness (ILI), is an infectious disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). The severe consequences of the Covid-19 pandemic highlighted the need for low-cost, passive assessment methods to detect infected patients early to restrain disease spread. We have researched and developed Machine

Learning models for detecting Covid-19 from disruptions in longitudinal physiological signs (such as heart rate and steps) collected passively from consumer-grade smart wearables, without relying on prior history or human-reported symptoms. In some cases, our proposed models are able to detect Covid-19 infection from smartwatch data several days before biological symptoms (such as coughing, sneezing) fully manifest.

Neurolingustic Assessment from Speech

Speech is an effective biomarker for evaluating neurological disorders, such as Traumatic Brain Injury (TBI), and mental health conditions. Speech production and communication difficulties are common manifestations of disability after TBI (2% of the population), whereas speech patterns such as low pitch and monotonous speech are effective indicators of depression (8.4% of the population). To alleviate healthcare burdens and reduce rehospitalization, passive speech monitoring through mobile devices offers a promising approach at scale, requiring minimal subject involvement while providing more accurate assessments compared to traditional methods that require active engagement and clinic visits. We have researched and developed AI models for passive neurolinguistics assessment of TBI and depression from speech. Our work addressed three major challenges in employing

Deep Neural Networks (DNN) for continuous paralinguistic health assessment on smartphones: energy efficiency, adverse recording environments, and speaker privacy

IMPACT Pain Center Grant

Chronic pain affects 50 million U.S. adults and interferes with the work or life of 25 million; however,despite effective treatments, patient response remains unpredictable. Mindfulness based interventions are effective at treating chronic musculoskeletal pain and Mindfulness-based Stress Reduction (MBSR) is recommended by the American College of Physicians as a first-line treatment for Chronic Low Back Pain (cLBP). Reliable biomarkers are urgently needed to predict and monitor treatment response to effectively target persons with cLBP with mindfulness interventions, especially diverse populations which are at increased risk of chronic pain and related health disparities.

To address this need, we are conducting research to identify biopsychosocial predictive and monitoring markers of the response of individuals with cLBP to MBSR. A robust combination of biopsychosocial factors that have been documented as strong potential candidates in pain and/or mindfulness intervention studies will be collected, including physical activity, depression, and social support. Our data analysis methodology will utilize Machine Learning (ML), a subfield of Artificial Intelligence, that enables discovering complex, multivariate relationships in multimodal data and building predictive models to extract comprehensive, accurate, and reliable biopsychosocial predictive and monitoring markers of the response to MBSR for cLBP in ethnically and racially diverse patients.