Geetika Singh

PRE-PROCESSING PIPELINE DEVELOPMENT FOR DBDP

• Obtaining data for analysis in the desired and right format is a tedious job.

• Built pre-processing functions for Digital Biomarker Pipeline Discovery (DBDP) to obtain pandas and/or .csv files for data from wearable sensors.

• The pipeline preprocess modifies column names, changes timestamps, calculates elapsed time and converts the file type to accessible .csv file for further analysis.

• The pipeline is developed using Python Programming languages. Libraries used pandas, matplotlib, numpy, seaborn, datetime, pytz, os, sys, json, rowingdata, mne, re

• Skills acquired : Writing pre-processing functions, Reading data from different file types like .tcx, .EDF, etc. and Blog writing

• https://github.com/DigitalBiomarkerDiscoveryPipeline/Pre-process/tree/master/pipeline

PREDICTION OF BLINDNESS IN GLAUCOMA PATIENTS

• Used data from Electronic Health Records and Durham Neighborhood Compass to predict blindness in glaucoma patients and analyze progression of blindness.

• The project involved analysis using R programming in the PACE environment. 

• Used multi-variate statistical model, random forest, support vector machine algorithm for analysis and performed survival analysis to study progression of blindness. 

• Skills acquired : Preprocessing data to create blindness labels using medical definition of blindness, exploratory data analysis in R, modeling, poster presentation.

CENTER OF MASS DETECTION OF LEFT VENTRICLE OF THE HEART

• Used Echocardiograms to detect the center of mass of the left ventricle of the heart. 

• The goal of the project was to find a pattern in change of center of mass.

• The class also provided the opportunity to understand heart anatomy through detailed dissection sessions and lectures in the Duke medical school.

• The analysis was done in MATLAB.

• Skills acquired : Image processing, Pig heart dissection, Algorithm development, Understanding 3D images, Using Philips QLabs software

MIMIC DATASET : MORTALITY PREDICTION

• The goal of the project was to predict mortality in patients being admitted in the hospital.

• Data was downloaded from Physionet MIMIC website and the data was accessed directly from the database.

• The project was implemented using Python programming. Libraries used : pandas, numpy, string, re, sql, sqlite3, sklearn, matplotlib

• Skills acquired : Accessing data from database, Exploratory data analysis, Data preprocessing, feature extraction, modeling (logistic regression, random forest classifier) using lasso.

PREDICTION OF ACTIVITY & CALORIES BURNT USING WEARABLES DATA

• Built a classification & regression model to predict activity and calories burnt respectively, using Apple Watch Dataset.

• The model was tested using Fitbit dataset.

• The type of activity was classified using classification algorithms - Support Vector machines and Decision Trees, whereas, the amount of calories burnt was predicted using regression models - Multiple Linear regression and Multivariate adaptive regression splines (MARS).

• The project was implemented in R using libraries like dplyr, tidyverse, caTools, e1071.

• Skills acquired : Data pre-processing, Classification (support vector machines and decision trees) and regression (multiple linear regression and multivariate adaptive regression splines) modeling, cross validation, test-train validation.

KAGGLE's WiDS DATATHON 2020

• The aim of the project was to create a model to predict patient survival using data from first 24 hours of intensive care.

• We achieved a Area under Receiver Operating Characteristic Curve of 0.88296.

• Skills learnt & Models used : Data imputation, Data Normalization, Generalized Boosted Regression Modeling (GBM), Lasso Regression, Logistic Regression, AUC score, Sensitivity & Specificity

COVID-19 ICU PHYSIOLOGICAL DATA ANALYSIS

• Nutritional and metabolic needs can result in under or over-feeding which can in turn result increased probability of mortality or problems after discharge.

• To understand metabolic profile of COVID patients using data collected using PhysioFlow (wearable device) and BIA devices from patients in the ICU.

• Determine effect of COVID on muscle physiology and body composition.

• Data for the study is obtained under the LEEP COVID study being conducted at Duke University.

• Analysis is being done using Python programming language. Libraries used till now : Pandas, Numpy, Plotly, Scikit-learn, Seaborn, Datetime, Missingno, Matplotlib

• Skills acquired : Extensive Exploratory Data Analysis, Development of Pre-processing functions for Physioflow and InBody BIA device data, Statistical decisions regarding outliers, Imputation and modeling, Research - oriented thinking

ASSESS ACCEPTABILITY OF DATA SHARING

• The goal of the project is to understand acceptability of devices from wearables and mobile phones (health apps) data sharing in Duke patients through surveys.

• The project is on-hold due to the COVID-19 pandemic and alternate methods of survey deployment is being considered.

• Skills acquired : Filing IRB application, Building survey questions in collaboration with behavioral scientists, doctors and professors, Using Qualtrics survey building resources.

FETAL HEAD CIRCUMFERENCE MEASUREMENT USING ULTRASOUND IMAGES

• Used Unet Segmentation algorithm and Convolutional Neural Network for segmentation and fetal head measurement respectively in TensorFlow.

• The goal of the project was to measure fetal head circumference to monitor fetal health during the gestation period.

• Ultrasound images were used from publicly available HC-18 dataset on the Grand Challenge website.

• Segmentation method using Unet was used to detect fetal head. Regression algorithm VGG-16 was used to calculate the fetal head circumference using ultrasound images.

• I used Python programming language and tensorflow to complete the project.

• Skills acquired : Use of tensorflow library, implementation of Unet segmentation algorithm, Convolutional neural network, VGG-16 algorithm and Resizing images.

INDIAN LIVER PATIENT DATASET ANALYSIS

• Obtained data from UCI Machine Learning Repository.

• Used unsupervised learning to cluster individuals with liver illness using R. The goal was to identify similarities in a set of patient population with liver disease, this could be used to further identify the type of liver disease.

• Used labels of the data to build a classification model to classify people with and without liver illnesses using MATLAB.

• The project was completed using R and MATLAB. Libraries such as : dplyr, tidyverse, Amelia, mlbench, Rcpp, Hmisc, lattice, caTools, e1071, 

• Skills acquired : K-means clustering, Principal component analysis, Preprocessing data in R, Exploratory data analysis in R, modeling (linear regression, support vector machine).

MIT COVID19 CHALLENGE - BEAT THE PANDEMIC II (HACKATHON)

• Ideated a non-invasive device to detect biomarkers to predict infection.

• The team consisted of 7 members and the project pitch was presented for the panel.