University of Pittsburgh researchers have developed a novel method to quantify the extent of pneumonia and other thoracic diseases from two-dimensional (2D) x-ray images, improving the diagnostic ability of such images.
Description
This deep learning technology, convolutional neural network (CNN), is used to develop a model to quantitively and accurately assess the extent of thoracic disease. These images are used in conjunction with corresponding three-dimensional (3D) computed tomography (CT) scans to generate ground truth for the machine learning paradigm. The approach has the ability to add useful quantitative data to 2D x-ray images and thus improving diagnostic tools.
Applications
• X-ray imaging
• Diagnostic tools
Advantages
Previously, it was extremely difficult to accurately quantify the disease extent from 2D images due to overlap of structures in various diseases. Thus, x-ray-based misdiagnosis in hospitals occurs frequently and costs patients, hospitals, and insurance companies in the long term. This technology provides a solution by detecting the disease and generating quantitative metrics automatically for physicians and providers. This software can be easily integrated into hospital cloud-based systems for point-of-care access and can even be used to trend disease progression between patient scans. Overall, by using CNN technology and generating ground truth from 3D images, this innovative method improves diagnostic performance of 2D images with heightened efficiency.
Invention Readiness
There is a prototype available, as well as experimental data providing preliminary evidence for the feasibility of the proposed technology for analyzing the extent of emphysema from 2D chest x-ray images in a cohort of COPD patients and pneumonia in a cohort of COVID-19 patients.
IP Status
https://patents.google.com/patent/US20220084193A1
