Spring 2022 Project Awards

CliniPane: Developing a “Third Paradigm” clinical intelligence application

Jonathan Handler, MD, FACEP, FAMIA, Senior Fellow for Innovation at OSF HealthCare
Omar Elabd, MCS, University of Illinois Urbana-Champaign
Roopa Foulger, OSF HealthCare
Marlene Granda, University of Illinois Urbana-Champaign

This proposal aims to design and, if feasible, build “CliniPane”, a system to serve as a sort of clinical HUD on the user’s desktop. CliniPane would sit alongside the EHR, interact in real-time with it to garner clinical context and content, receive formatted information relevant to that context from a clinical intelligence engine, and non-interruptively push that information to the clinician via the application’s interface. Feedback provided by the user (volitional and potentially passive) through the interface will allow the application to continuously learn user preferences to optimize the user experience. In sum, the aim is clinical decision support that is ambient, anticipatorily pushed, needs-based designed, and continually learning.

Enhanced Focality of Transcranial Magnetic Stimulation Using an Ultrathin Wearable Metasurface for Treating Neurological Disorders

Yang Zaho, PhD, University of Illinois Urbana-Champaign
Yun-Sheng Chen, PhD, University of Illinois Urbana-Champaign
Huan Huynh, MD, OSF HealthCare

The objective of this proposal is to demonstrate the feasibility of an innovative ultrathin metasurface device that is compatible with TMS to enhance the spatial resolution and control the focal distribution of the stimulating field. To attain the objective, the investigative team aims to:

Design, fabricate, and test a metasurface device with a controllable surface impedance that reshapes the magnetic fields from the stimulator and thus to generate a refined focus; and
Incorporate an integrable dual-modal photoacoustic/ultrasound imaging technology that images the brain oxygen saturation and blood flow, which are known to vary during TMS.

The dual-modal imaging technology can be seamlessly integrated and used together with electroencephalogram (EEG) as real-time physiological feedback for demonstrating the metasurface-enhanced focus during TMS.

VRtual Ed: A Virtual Three-Deminstional Educational Platform for Healthcare Students

Avinash Gupta, PhD, University of Illinois Urbana-Champaign
Lydia Lee, University of Illinois Urbana-Champaign
John Shallat, MD, OSF HealthCare
Maureen Matthews, APRN, OSF HealthCare
Samantha Bothwell, MSN, RN, University of Illinois Chicago
Celeste Schultz, PhD, University of Illinois Chicago

The proposed project will develop a virtual 3D interactive education platform using the Oculus Quest 2 as the head-mounted device (HMD) and Unity gaming software to create the hospital environment and interactive human-like digital avatars (created with the Reallusion Character Creator) integrated with artificial intelligence (Microsoft Language Understanding (LUIS), Google Speech Recognition) to replicate real-world interactions so that nursing students can gain the skills to become competent caregivers.

CAnPredict: An Algorithm for Improved Pancreatic Ductal Adenocarcinoma Detection

Sonia Orcutt, MD, Department of Surgery, University of Illinois College of Medicine at Peoria
Ravishanakar K. Iyer, PhD, University of Illinois Urbana-Champaign
Christopher Gondi, PhD, University of Illinois Cancer Center
Lusine Demirkhanyan, PhD, University of Illinois College of Medicne Peoria
Andrew Darr, PhD, University of Illinois Urbana-Champaign
James Weldy, OSF HealthCare
Nathan Pritzker, MBA, OSF HealthCare
Mosbah Aouad, University of Illinois Urbana-Champaign

The overarching goal of this proposal is to enhance the early detectability of PDAC and hence improve patient survival rate. To do so, our interdisciplinary research team, with expertise in basic and clinical science, machine learning (ML) and artificial intelligence (AI), will work together to develop a predictive diagnostic algorithm based on existing patient historical, multimodal data. While ML approaches have been applied to other disease models, there is no known example of successful prediction of disease onset (2-6) , and to the best of our knowledge, no such diagnostic tool is available for expedited prediction and detection of pancreatic cancer. This proposal seeks to bridge the gap in existing diagnostic models to achieve an early, accurate and a predictable outcome with the ultimate goal of improving patient survival. Successful completion of the work will not only be relevant to early pancreatic cancer diagnosis, but also establish new ways to perform diagnostic targeting of several other cryptic disease processes.

Incentivization of Health Care Initiatives with Cryptocurrency in a Zero-Knowledge System

Jonathan Handler, MD, FACEP, FAMIA, Senior Fellow for Innovation at OSF HealthCare
Tate Ralph, OSF HealthCare
Wencui Han, PhD, University of Illinois Urbana-Champaign
Andrew Miller, University of Illinois Urbana-Champaign

The current proposal seeks to implement a decentralized application utilizing blockchain technology to overcome this obstacle by employing the methodology of zero knowledge proof to provide double anonymity in the incentivization transaction. In this proposed architecture, patients would provide compliance data to the application and receive compensation based on the validity of their compliance. Health systems would provide the remuneration through an anonymous matching of the patient to a medical record within their electronic medical record. The patient’s ID would remain unknown, but the fact that the patient ID existed in the health system would be validated through implementation of zero knowledge proofs. The particular health system that matches to the patient ID is irrelevant to the value proposition for the patient and does not need to be exposed. Therefore, with both the patient remaining unidentified to the health system and the health system remaining unknown to the patient, double anonymity is achieved. With double anonymity, the claim that a health system is remunerating for compliance specifically to retain use of its services loses substance. No direct connection between the patient and health system can be identified. If successful, this proposal has the potential for impact, as direct compensation is likely the most effective tool for driving compliance and, therefore, improvements in individual and population health.

Toward Machine Learned Aortic Arch Measured Diameters

Matthew Bramlet, M.D., Department of Pediatrics University of Illinois College of Medicine – Peoria; Advanced Imaging and Modeling Lab, Jump Simulation
Brad Sutton, PhD, University of Illinois Urbana-Champaign

This project seeks to create a new tool that allows each institution to generate their own normative data which is relevant to their own institution, modality and methodology. In doing so, it will create confidence in clinical decisions based on local measures and experience. Current methods of generating pediatric normative data are cumbersome and time consuming. Automated methods of generating normative data have not yet been developed.

Development of a Chatbot for Delivering Long-Term Motivational Interviewing for Improving Exercise Adherence in Hemodialysis Patients

Jessie Chin, PhD, School of Information Sciences, University of Illinois at Urbana-Champaign
Suma Bhat, University of Illinois Urbana-Champaign
Ben Pflederer, MD, OSF HealthCare
Chung-Yi Chiu, PhD, University of Illinois Urbana-Champaign
Ken Wilund, PhD, University of Illinois Urbana-Champaign
Rehan Shah, University of Illinois Urbana-Champaign

Given the compromised health conditions and high self-management demands, hemodialysis patients have difficulty adhering to exercise programs. Motivational Interviewing (MI) is a counseling approach demonstrated to promote positive behavior change but is often impractical to conduct because of the administration costs (e.g., qualified professionals or frequency of sessions required). A possible alternative would be to develop an automated conversational agent to deliver the MI, but no prior research has demonstrated the feasibility or effectiveness of delivering long-term MI using natural conversations. The proposed study will bridge theories in behavioral sciences (e.g., MI and stages of behavioral change) and Natural Language Processing (NLP) to develop a long-term MI conversational agent, LogMintBot, for exercise adherence.

Pediatric Automated Intelligent Respiratory Support (PAIRS): Development of an Automatic Oxygen and Flow Weaning System for Pediatrics

Keith Hanson, MD, PhD, Department of Pediatrics, University of Illinois College of Medicine Peoria
Ramavarapu Sreenivas, PhD, University of Illinois Urbana-Champaign
Adam Cross, MD, OSF HealthCare
Roopa Foulger, OSF HealthCare
Jonathan Gehlbach, MD, OSF HealthCare

The Pediatric Automated Intelligent Respiratory Support (PAIRS) system is an automated weaning system that will meet this need. The system will take into account all of a patient’s vital signs and adjust respiratory support as the patient is recovering. The software of the PAIRS system connects to and manipulates a modified Heated high flow nasal cannula device so that it can titrate both the FiO2 and the flow rate independently. This system can wean more frequently, and with less subjectivity, compared to human providers, whose assessments represent only single points in time and are biased based on experience and other factors. In contrast to typical clinical practice, the PAIRS system provides continuous patient assessment and real time intervention. The PAIRS system is hypothesized to improve care and decrease length of stay for pediatric patients who require respiratory support. The current proposal is to develop a prototype system with all the key components and test it in a simulated environment. Future work will include incorporating machine learning to improve the system logic and testing in the clinical environment.

Development of a pneumothorax computational model toward lung metastasis visualization and modeling

Matthew Bramlet, M.D., Department of Pediatrics University of Illinois College of Medicine – Peoria; Advanced Imaging and Modeling Lab, Jump Simulation
Brad Sutton, PhD, University of Illinois Urbana-Champaign
Dan Robertson, MD, OSF HealthCare
Alexa Waltz, OSF HealthCare
Olivia Bryan, OSF HealthCare

This project, seeks to build on the utility of virtual reality based 3D modeling of pre-surgical anatomy by expanding beyond our previous work with congenital heart disease and large tumor resection which have already been fully implemented within the Children’s Hospital of Illinois (CHOI). At the core of this impact is the ability to recreate the surgical field in full 3D prior to the actual procedure allowing the surgeon to embed a more accurate mental representation of the surgical field in their mind prior to surgery. The more accurate 3D mental representation allows for more accurate surgical planning and decreases time under anesthesia “getting oriented” to the surgical anatomy as well as freeing working memory that would otherwise be spent aligning the reality of the surgical field to the 2D image surrogates of the anatomy. Surgical 3D localization of lung metastases is very difficult and affords much opportunity for direct translational patient impact. While 3D segmentation and modeling of lung lesions is possible within CHOI’s existing capability, the 3D surgical localization changes significantly with any deflation of the lung. This project looks to bridge the gap between existing capability and computational modeling of 3D localization in a deflated lung toward improved 3D mental representation of actual deflated surgical lung lesions.

Koopman framework for detecting mental health changes in multimodal wearable data

Manuel E. Hernandez, PhD, College of Applied Health Sciences, University of Illinois at Urbana-Champaign
Jean Clore, PhD, University of Illinois Urbana-Champaign
Richard Sowers, PhD, University of Illinois Urbana-Champaign
Elizabeth Hsiao-Wecksler, PhD, University of Illinois Urbana-Champaign

This interdisciplinary project aims to intuitively and intelligently collect, sense, connect, analyze and interpret wearable data from multimodal sensor systems to enable discovery of mental health symptoms, such as anxiety, and optimize health in adults. This objective will combine development of an ML/AI framework for detecting and predicting short-term and long-term changes in state anxiety, with the use of multimodal sensors to collect electrophysiological, acoustic, and/or kinematic measurements, and well-established psychosocial paradigms. Excessive anxiety has been shown to have detrimental effects on physical and cognitive performance. While self-reported measures have been used as a gold standard for evaluating anxiety, they are not feasible for continuous monitoring of anxiety and are unable to provide a measure of real-time, event-contingent changes in anxiety. Remote monitoring tools can provide objective and continuous monitoring and prediction of anxiety in vulnerable populations. However, biosensor integration with data analytic approaches are lacking. Collected physiological data often exhibit multiple complicating factors, e.g. noise or sparse and irregular sampling, due to biofouling of sensing platforms and/or complexity of physiological data. Fundamental knowledge connecting sensors with existing models and/or ML is missing.

Advanced Auscultation Audio Algorithmic Analysis (a⁵)

Adam Cross, MD, FAAP, Clinical Research Informaticist, OSF HealthCare
Jennifer Amos, PhD, University of Illinois Urbana-Champaign
Eliot Bethke, University of Illinois Urbana-Champaign

The proposed project will focus on advanced feature extraction and processing to improve analytical performance to enable end-to-end explainable output and avoid the “black box” problem so prevalent among current models in the literature. The project will use the publicly available ICBHI 2017 dataset of 920 lung audio recordings from 126 subjects. We will implement raters trained in auscultation to label adventitious lung sounds as well as inhalation, exhalation, and other notable lung sounds in our dataset, then use this data to build and test the algorithm. The output of the model will be critically reviewed by human experts, taking careful note of lung sound types and diagnoses that are challenging to label for human raters and for the algorithm.