Sitemap

CDS Faculty Fellow Elena Sizikova’s CURP Team and Colleagues Create an Automatic TB Analysis Algorithm

3 min readOct 20, 2021

It is critical for medical imaging to be able to accurately diagnose and evaluate infections. Computed tomography (CT) scans are most effective at detecting and assessing pulmonary infections but due to the expensive cost of CT scanners, they are less available in lower-resource and smaller or rural communities and areas. Conversely, X-rays, a different type of imaging process, are affordable, more widely available, but provide simpler, two dimensional images that may not capture as many details. Taking a look specifically at tuberculosis (TB), CURP (CDS Undergraduate Research Program) scholars Ashia Lewis (The University of Alabama) and Evanjelin Mahmoodi (UCSC), CDS MS student Yuyue Zhou, CDS Faculty Fellow Elena Sizikova, and Professor Megan Coffee (NYU Langone) demonstrated the benefits of relying on synthetically generated CT scans in their recent paper “Improving Tuberculosis (TB) Prediction using Synthetically Generated Computed Tomography (CT) Images”. The paper was accepted into ICCV Workshop on Computer Vision for Automated Medical Diagnosis (CVAMD) 2021, which took place October 11–17, 2021. In their research, the team leverages an existing computational model, X2CT-GAN (Ying et al., CVPR 2019), a neural network trained to predict CT from chest radiographs (X-ray). The team showed that using synthetically generated CT images, one can improve the accuracy of an automatic classification model trained on X-rays only. Additionally, they analyze the ways in which different types of inputs influence how accurately TB is identified and disease properties are classified. Their findings indicate that “synthetically generated CT improves TB identification by 7.50% and distinguishes TB properties up to 12.16% better than the X-ray baseline.”

To the best of their knowledge, the team is the first to use synthetically-generated CT images for the analysis of TB. It is their hope that their work will inspire further study into how synthetic data can be used to improve automatic disease diagnosis. The proposed approach allows for automatic detection but also allows healthcare providers access to technology they may be cut off from. CT scans can provide doctors and nurses with more information than they have been able to have before.

About the Team:

Ashia Lewis is a Spring 2021 CURP (CDS Undergraduate Research Program) scholar, advised by Dr. Elena Sizikova and Professor Megan Coffee. She is also currently a student at the University of Alabama.

Evanjelin Mahmoodi is a Spring 2021 CURP (CDS Undergraduate Research Program) scholar, advised by Dr. Elena Sizikova and Professor Megan Coffee. She is currently a student at the University of California, Santa Cruz.

Yuyue Zhou is a CDS alumna, holding an MS degree in data science.

Megan Coffee is MD, PhD is Clinical Assistant Professor at the Department of Medicine, NYU Langone.

Elena Sizikova is a CDS Faculty Fellow and holds a PhD in the Graphics/Vision Lab in Princeton’s Computer Science department.

By Ashley C. McDonald

--

--

NYU Center for Data Science
NYU Center for Data Science

Written by NYU Center for Data Science

Official account of the Center for Data Science at NYU, home of the Undergraduate, Master’s, and Ph.D. programs in Data Science.

No responses yet