Online Program

Return to main conference page
Friday, May 31
Data Visualization
Data Visualization in Applications
Fri, May 31, 10:30 AM - 12:05 PM
Regency Ballroom EF

Topological Data Analysis for Understanding Phenotypic Presentation in Aortic Stenosis (305175)

Grace Casaclang-Verzosa, West Virginia University 
Partho P Sengupta, West Virginia University 
*Sirish Shrestha, West Virginia University 

Keywords: topological data analysis, network analysis, network visualization

Echocardiographic assessment of left ventricular (LV) systolic and diastolic function is an integral part of evaluating patients with subclinical or overt cardiac function. A standardized echocardiographic protocol can produce numerous images and parameters for assessing hemodynamics and structural features of the heart. Due to the large dimensions produced during the assessment, integrating the data to ascertain valuable clinical deduction is difficult. Network biology may present to be a powerful model for interpreting and contextualizing large and diverse sets of biological data. Topological Data Analysis (TDA), a framework for machine learning, may provide a valuable network visualization of similar and dissimilar patients for extracting novel relationships in complex and heterogenous dataset. It represents the complex and high-dimensional data into a compressed and interactive network by grouping similar patients into nodes and connecting overlapping nodes by edges that describes the shape of the data. The subsequent generation of the disease map permits exploration of multi-dimensional complex data such as aortic stenosis (AS) and may allow clinicians to track the patients within the disease map as their valvular dysfunction worsens. In our study of AS, TDA was applied to discern patient similarity for phenotypic recognition of LV responses during the progression of AS. It formed a loop structure in which the two spectra of the disease severity were separated at two opposite ends of the map while linked through the moderate AS at the top and the bottom of the loop. Interestingly, TDA was also able to capture the patients’ improvement following the intervention and show their regression to moderate or mild AS. The network was validated against the mice data that resulted in clustering of mice that closely paralleled human data. TDA may be a viable methodology for predicting disease progression, generating focused patient treatment, and assessing prognosis.