Spring 2019 Projects

Corning: Hyperspectral Imaging To Determine Microcirculatory Function In Patients With Sepsis

Team Members: Fawzi Ali, Joe Buckley, Vicky Li, Mcvvina Lin

Hyperspectral imaging (HSI) is a spectral imaging acquisition technique where each pixel of an image represents the information within certain spectral bands. In this project, HSI data was used to capture microcirculatory stress on the palm of a human hand. The goal of this project was to determine if HSI can be utilized as non-invasive method of measuring microcirculatory stress. Team members built correlation matrices and tracked trends, identifying potential patterns between HSI bands. They also used SVM, PCA, and random forest models to predict blood sample outcomes.

Hajim School of Engineering Advising Office, University of Rochester

Team 1: Predict Student Outcomes in College Courses

Members: Mackenzie Lee, Ji Eun Han, Abraham Do

The goal of this project is to identify correlations between high school-level academic preparedness, performance in introductory math and physics courses, and performance in students’ major specific courses. Team members used random forest and multivariate linear regression models to discern patterns in the data.

Team 2: Identify Students at Risk of Probation Using Predictive Models

Members: Zijian Song, Elliot Frost, Chen Ye

The goal of this project is to determine which students are at risk of academic probation and, more importantly, how to prevent it. Members built logistic regression-based models to identify factors that commonly lead to academic probation.

Origent: Optimize Disease Progression Rates in Lou Gehrig Disease (ALS)

Team Members: Xinxin Gu, He Huang, Yonghao Duan, Yuexi Wang

Many drugs used to treat amyotrophic lateral sclerosis (ALS) have been tested in clinical trials (trials are key to understanding the behavior of new drugs). The Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS) is an instrument for evaluating the functional status of patients with ALS. If we can predict the ALSFRS-R score accurately and without intervention, we can use these predictions to evaluate the efficacy of new ALS drugs. The goal of this project is to predict, given specific time t, the estimated ALSFRS-R score for a new patient, determining which patients would most benefit from inclusion in the clinical trial. Modeling tools used in this analysis included Neural Network, Random Forest, XGBoost, and the Weighted Ensemble method.

Ortho Clinical: Test Failure Prediction for In-Vitro Diagnostic Equipment

Team Members: Michael Sklar, Sam Rusoff, Fengyi Zhao, Yankun Gao

OGR test results are crucial for evaluating the performance of diagnostic clinical testing devices. Early warning for erroneous (increased) OGR values allows device makers to improve their customers’ experiences as quickly as possible. The goal of this project is to build machine learning models that can predict OGR rates and evaluate the attributes that drive increases in OGR results. Team members utilized Logistic Regression and Random Forest models to predict OGR ratios.

URMC: Predict Accidental Falls in At-Risk Cancer Patients

Team Members: Boyu Liu, Sixu Meng, Junchao Shen, Zhikang Jiang

Accidental falls are common in aging populations, but falls have not yet been widely studied for aging cancer patients. Students analyzed clinical and patient questionnaire data to determine the correlation between predictor variables related to cognitive function, motor function, cancer type, social behavior, and fall rate. The team used machine learning methods such as Recursive Feature Elimination to select features, then combined them with logistic regression to predict fall rates.

Vnomics: Automatic Configuration of Vehicle Drivetrain

Team Members: Jiayin Han, Yutong He, Chuangyu Lou, Zetian Xiao

Manual system configuration is error prone and slows down the installation process of Vnomics’ True Fuel product. The goal of this project is to determine the number of gears and associated gear ratios (slope of engine versus wheel speed) in a vehicle’s transmission, using sensor information associated with throttle position, wheel speed, and engine speed (RPM). Our system automatically identifies the number of gears and the slope of each gear, increasing the automation of truck configuration during the installation process. Vnomics can then use the system to improve the accuracy and speed of True Fuel installation. To create it, team members used convolutional neural networks (CNN) to predict the number of gears in each vehicle. Then they used the expectation-maximization (EM) algorithm to predict the slopes of the gears.