Individual CDI Risk Forecasting

Mentors: Dr. Bijaya Adhikari, Dr. Sriram Pemmaraju
Domain Expert: Dr. Philip Polgreen

Clostridioides difficile infection (CDI) is a bacterial hospital acquired infection (HAI). HAIs like CDI are a major burden on both patients and healthcare professionals (HCPs). The CDC estimates that there are roughly 4.5 HAI cases for every 100 hospital admissions with an annual cost of between 28 and 45 billion USD. Timely forecasts of HAI risk can help inform the deployment of effective counter measures, as well as improve resource allocation planning by forecasting future ICU bed requirements.
In this project, we will develop AI-driven approach to forecast daily CDI infection risk for each patient. Our proposed approach will combine both CDI exposure and patient demographic information to estimate the patient’s current risk for CDI infection.

Personalizing Hearing-Aids

Mentors: Dr. Octav Chipara
Domain Expert: Dr. Yu-Hsiang Wu

Approximately 35 - 50% of Americans over 65 years old report having an age-related hearing impairment treated primarily with hearing aids. Regular use of hearing aids has been shown to improve communication and avoid the harmful effects of hearing loss, including an increased risk of social isolation and depression. Over the last decade, researchers have successfully improved the performance of hearing aids by leveraging the increasing capabilities of hardware platforms, which enabled the deployment of sophisticated signal processing algorithms such as those for noise suppression, directional microphones, and auditory scene classification. In contrast, future internet-enabled hearing aids will overcome limited hardware resources by accessing cloud services and edge devices' significant computational and storage capabilities. As part of this development project, you will learn how to design, train, and deploy machine learning algorithms that enhance the capabilities of hearing aids. These algorithms will run either in the cloud or on resource-constrained edge devices.

Modeling Problems Related to Spread of Hospital Acquired Infections

Mentors: Dr. Alberto Maria Segre
Domain Expert: Dr. Philip Polgreen

Students will work on modeling problems related to the spread of hospital acquired infections as part of the Computational Epidemiology research group. Team method is data driven, based on fine-grained data obtained from the University of Iowa Hospitals and Clinics and other similar healthcare facilities. Our approach generally involves running simulations to measure the effectiveness of changes to hospital architecture, hospital operations, healthcare worker behaviors (e.g., hand hygiene), and patient room assignments. Thus, depending on student interests, research topics might include, for example:
1. Constructing generative statistical models of, e.g., healthcare worker behavior from large, fine-grained, healthcare worker movement data sets;
2. Constructing simulations of different hospital operation policies (e.g., room assignments, cleaning practices, etc.);
3. Extracting spatial models of healthcare facilities from CAD files; or
4. Devising statistical models of healthcare acquired disease transmission parameters (e.g., shedding rates, healthcare worker/patient contact parameters, etc.).

Computational Supports for Children’s Development of Executive Functions

Mentor: Dr. Juan Pablo Hourcade

We are developing interactive technology to lower barriers to a specific set of evidence-based activities intended to enhance preschool children's executive functions through high-quality social play. Our interactive web application, called StoryCarnival, includes interactive stories to motivate play, a play-planning tool, and voice agents controlled by adult facilitators to keep children engaged in play. We will expand our current research on StoryCarnival in multiple ways that we expect to contribute to the field of human-computer interaction. First, we are researching advanced user interfaces, including smart recommendations, for adult facilitators to control tangible voice agents with which children interact as part of StoryCarnival activities. Second, we will expand our focus from preschool children to neurodiverse children (e.g., autistic children, children with Down Syndrome).

Using Smart Glasses and Watches to Access Hearing Aids

Mentors: Dr. Octav Chipara
Domain Expert: Dr. Yu-Hsiang Wu

For any hearing aids to work, they must do the right things (e.g., enable directional microphones), for the right people, in the right places (when speech and noise are spatially separated). To this end, hearing aids must accurately identify challenging and critical listening situations, the characteristics of these situations, and the listening goals of listeners. One commonly used method to assess the performance of hearing aids in the real world is to use Ecological Momentary Assessment (EMA) self-reports. EMA is a methodology that asks respondents to repeatedly report their experiences during or shortly after the experiences in their natural environments. Although EMA in audiology research is increasing, EMA is not a tool without limitations. For example, it is generally tricky for EMA to capture challenging listening situations when EMA surveys are delivered at random as they occur infrequently. In addition, the association between discrete EMA data and continuous sensor data (i.e., how listeners integrate the experience of the past several minutes to answer an EMA survey question) is unknown. This project aims to use smartwatches and smart glasses to (1) capture more challenging listening situations with higher fidelity than previously possible and (2) determine the association between EMA self-reports and continuous sensor data is using novel machine learning algorithms.