Big Data on Healthcare Application

Permanent URI for this collection


Recent Submissions

Now showing 1 - 4 of 4
  • Item
    Risk Stratification and Prediction of Postoperative Complications Using Temperature Trajectories
    ( 2022-01-04) Padman, Rema ; Grant, Jennifer ; Ravichandran, Urmila ; Bashyal, Ravi ; Shah, Nirav
    Early identification of patients at highest risk of postoperative complications can facilitate appropriate diagnostic work-ups and earlier interventions. We investigate whether postoperative temperature trajectories can stratify patients and predict this risk via a retrospective study of 5,084 adult patients undergoing elective primary total knee arthroplasty at a major health system. Demographics, surgery duration, temperature readings, length of stay, comorbidities and complications were extracted from the data warehouse. Group-based trajectory modeling was applied to cluster patients into distinct groups following similar progression of maximum temperature over four-hour time intervals until discharge, and group information was included in predicting risk of critical complications. Three non-overlapping, temperature-based trajectories were identified as high- (8% of patients), medium- (49%), and low-risk (43%) groups. Complication rates were significantly higher in the high-risk group (16.7%), than the medium-risk (5.4%), and the low-risk groups (2.70%) (p<0.01). Group information shows promise in improving complication risk prediction for high-risk patients.
  • Item
    Automating Blood Flow Simulation Through the Aorta in Patient-specific CT Images
    ( 2022-01-04) Habijan, Marija ; Galic, Irena
    Computational fluid dynamics (CFD) modeling of blood flow is significant for obtaining patient-specific hemodynamics information for functional assessment of the cardiovascular system. In this work, we present a framework for fully automatic CFD simulation through the aorta. The proposed framework consists of four main stages: (1) automatic segmentation of the aorta, (2) model generation, (3) mesh creation, and (4) blood flow simulation. In the segmentation part, we utilized a 3D MultiResUnet network for automatic segmentation of organs at risk from the CodaLab SegThor Challenge. After that, we extract ascending and descending aorta and further proceed with the model and mesh generation. Finally, we simulate the pressure along the surface of the aorta, the displacement, and the velocity. The entire framework was implemented in Python with open-sourced dependencies (Pytorch, VTK, SimVascular, SimpleITK), can be executed from the command line, and does not require user intervention, significantly reducing aorta simulation time.
  • Item
    An End-to-End Machine Learning Solution for Anxiety and Depressive Disorder Symptom Occurrence During COVID-19: A New York Case Study
    ( 2022-01-04) Yadav, Nikhil ; Singh, Christopher ; Srivastava, Kajal
    Anxiety and depression during the COVID-19 pandemic have heightened as evidenced by the rapidly growing corpus of research articles suggesting a link between the pandemic and mental health. This paper proposes a unique end-to-end user-centric machine learning (ML) architecture, capable of assessing the quality of ML predictions about the occurrence of anxiety and/or depression symptoms. A case study is presented using official New York State COVID-19 data, highlighting the plug-and-play capabilities of this architecture for both external features, and newer ML models. This is demonstrated through the formal design of a custom weighted clustering algorithm which outperforms conventional unsupervised techniques in grouping symptomatic cases. The ability to augment external sentiment data mined from social media platforms like Twitter, increases the predictive power of this architecture. This work serves as a blueprint to build a practical ML solution to better gauge the effect of future pandemic waves on mental health.
  • Item
    Introduction to the Minitrack on Big Data on Healthcare Application
    ( 2022-01-04) Tsoi, Kelvin ; Hung, Patrick ; Poon, Simon