Non-invasive Measurement of Pulmonary Hypertension in Premature Infants

Overview

The Penn State Hershey College of Medicine wants to develop a technology for imaging the pulmonary vasculature for neonatal patients. Our task was to develop a new method for diagnosing pulmonary hypertension that is non-invasive but more reliable than existing non-invasive methods.

Objectives

Our objective was to generate a method for the diagnosis of pulmonary hypertension in infants born premature. Tackling this requires a multi-faceted approach: developing a phantom of the diseased and healthy states, classification of ultrasound images, and modeling wearable sensors.

Approach

– Conduct meetings with experts as needed to understand the patient and stakeholder needs

– Perform an extensive literature review on perfusion phantoms, tissue mimics, existing non-invasive imaging technologies, and image classification.

– Develop a perfusion phantom from an adult patient CT scans

– Use a handheld ultrasound probe to image the perfusion phantom

– Develop a convolutional neural network for image analysis and classification

– Classify the ultrasound images into the two states, then feed the images into the CNN

– Model wearable sensors on pulmonary regions of interest for future research

Outcomes

– The team developed a novel approach to diagnosing pulmonary hypertension using ultrasound images inputted to a neural network to classify whether a patient is diseased or not

– A perfusion phantom was successfully created to model the fluid dynamics and biological parameters of healthy and diseased patients

– When the phantom was imaged using color ultrasound there were noticeable differences in the diseased versus healthy state

– A neural network was architected so ultrasound images from the phantom could be used to train the network to distinguish between health and diseased images