PICTURE: Predicting Intensive Care Transfers and other UnfoReseen Events

Early detection of patient deterioration has been found to lead to reduced mortality risk, reduced length-of-stay and decreased hospital costs, yet identifying patient deterioration is a challenge for clinicians. PICTURE’s General Floor Analytic for adult and pediatric populations is a combination of machine learning algorithms utilizing electronic health record (EHR) data to passively and accurately predict ICU transfer or death as a proxy for patient deterioration. 

For further information, visit here for the adult model and here for the pediatric model.

Model Performance Diagnostic Suite (MPD)

The Model Performance Diagnostics  detect data shift and model degradation by comparing the distribution of the real-time data that is fed to the model to the data on which the model was trained.  MPD also estimates the performance of the model post-deployment, a unique feature that is currently not offered by any other tool, all via an interactive visual interface.

For further information, visit here.

Environment for Model Maintenance, Integration and Tuning (EMMIT)

Integrating machine learning models and predictive analytics with a hospital environment is difficult.  Data flowing in and out of the model can become a black box with no way to monitor if the models are operating or data transfer is effective.  Our Data Operations Team has designed EMMIT, a platform for hosting any number of unique models and helping to deploy them within an EHR system.  EMMIT also provides operational monitoring (uptime, alerts, error reporting)  to teams who would otherwise have no insight on the status of their software. 

For further information, visit here.

Prediction of Heart Failure Onset using Multimodal Data Analysis, Deep Learning and Commercial Wearables

The aim of this project is to apply machine learning and signal processing techniques to multiple types of data (electronic health records, electrocardiography, and heart rate variability) to predict heart failure onset, and to use wearable devices to improve the predictions and expand their impact to a wider population. The resulting tools will help prevent delayed diagnosis of heart failure, leading to improved medical outcomes, enhanced patient experience, and reduction in healthcare costs.

DETECT-ARDS - Analytic for Detecting Acute Respiratory Distress Syndrome

DETECT-ARDS is a new approach for identifying ARDS findings on chest x-rays.  With ARDS often missed or under-diagnosed, DETECT-ARDS has the potential to transform patient outcomes for the better.  In collaboration with Dr. Michael Sjoding, our team has trained a powerful deep convolutional neural network model that can identify findings consistent with ARDS with high accuracy.

For further information, visit here.

Breath Analysis Device for Disease Detection

This automated, portable breath analysis device utilizes gas chromatography and a corresponding algorithm to detect and monitor breath biomarkers indicative of Acute Respiratory Distress Syndrome (ARDS), COVID-19, and other lung injury.  Identifying the onset of these illnesses early and monitoring their trajectory over time can help stratify patients and better allocate resources within the hospital. The Ansari Lab, in collaboration with Dr. Xudong Fan's team from Biomedical Engineering, has been designing the automated algorithms to analyze the signals that are generated by the device.

For further information, visit here.

Wearable Body Odor Sensing for Disease Detection and Monitoring

Many diseases, both internal and cutaneous, have distinct odors associated with them, and their identification can provide unique diagnostic clues, guide laboratory evaluation, and facilitate and expedite treatment. Current body odor analysis relies on benchtop instruments, but they are too bulky for use at point-of-care, home or workplace. E-nose technologies provide a simple, light, and low cost alternative for body odor analysis, but they are highly susceptible to environmental changes (e.g., temperature and humidity). In collaboration with Dr. Xudong Fan's team, the Ansari Lab has been developing analytical tools to process the data that is generated by the sensors and use them to predict 20 different medical conditions.

Vascular Tone Monitoring System (VATMOS)

By continuously assessing peripheral vascular tone, this piezoelectric-optical wearable ring provides real time insights into how the cardiovascular system is responding to illness or injury before changes in traditional vital signs like blood pressure occur. This sensor provides almost immediate feedback to clinicians on the physiologic response to treatment. This work has been a collaboration with Dr. Kenn Oldham in Mechanical Engineering.

For further information, visit here.

Implementation of an All-Cause Deterioration Model for Adult and Pediatric Hospitalized Patients

The main objective of this project is to reduce patients' risk of experiencing adverse events during their hospital stay by deploying our PICTURE early warning system. To achieve this, we will create an efficient and effective user interface to deliver the PICTURE risk scores and their explanations to providers, and we will ensure that the clinical decision support system and its interface fit within their existing clinical workflow. 

For further information, visit here.

Clinical Implementation and Pilot Randomized Controlled Trial of PICTURE-Pediatric (Predicting Intensive Care Transfers and UnfoReseen Events)

The goal of this project is to design a clinical response for PICTURE-Pediatric, a pediatric deterioration index, using human-centered design. In addition, the team will design a pilot randomized controlled trial to test the usability of the model and its clinical response, as well as the feasibility of a larger multi-center RCT.

For further information, visit here.

Hypertrophic Cardiomyopathy Detection using Federated Learning

The main objective of this project is to develop a model for detection of hypertrophic cardiomyopathy using ECG signals and Echocardiograms. The Ansari Lab is collaborating with the American Heart Association to develop this model using federated learning.

CPR Artifact Removal in ECG Signals 

The main objective of this project is to improve, validate and deploy an existing model for removing CPR artifacts from ECG signals during chest compressions. This will allow for continuous CPR and eliminates the need to interrupt chest compressions to check the rhythm in order to make AED shock decisions. This work is a collaboration with the University of Connecticut and the Ansari Lab will focus on the validation of the model and development of a framework to run the model in hospital in real-time.

Optimizing Emergency Department Triage Decisions and Inpatient Bed Assignment 

The first objective of this project is to develop a recommendation model to improve triage decisions for ED patients, i.e., determine which patients will benefit from admission to the ICU versus a general hospital ward. The second objective of this project is to develop an optimization model to suggest the best assignment of ED boarded patients (patients admitted to the hospital from the ED but awaiting a bed) to available general ward beds.

Improving Management of Acute Decompensated Heart Failure (ADHF) using Reinforcement Learning

Management of ADHF is challenging due to the complex interplay between volume status, renal function, and blood pressure, each affected in complex ways by diuretics and goal-directed medical therapy. The primary objective of this project is to improve management of AHDF by recommending optimal treatment decisions personalized to patient characteristics using reinforcement learning.