Technology Description

Regional measures of lung function used for assessing chronic obstructive pulmonary disease (COPD) are typically computed by registering chest computed tomography (CT) scans acquired at different volumes, typically end-inspiration and end-expiration volumes. Since this process inherently requires CT scans at different volumes, it is not possible to estimate the local measures of lung function for patients where multiple volume scans are unavailable or are impossible to acquire. Another disadvantage of acquiring multiple CT scans is the exposure to radiation dosage which is a risk factor for cancer.
The researchers at the University of Iowa developed a generative learning framework that can be used to directly convert a CT scan at expiration either directly to an inspiratory scan or to a Jacobian image. The system architecture comprises two different generative neural network architectures that are used to generate these images using a single scan. Initially, the system is trained by ground truth Jacobians or TLCs to be generated by the generator. Once the system training is complete, the models can now be used to generate these local measures of lung function.
This technology can be used to rapidly characterize COPD, lung cancer, lung transplant, or other pulmonary disorders where dual volume CT scans are either not possible or not recommended. This is expected to have significant clinical impact where the proposed process would eliminate the need for multiple CT scans for the same subject.

• • It relies on a single scan from each subject, rather than multiple scans required by conventional approaches.
  • It results in less exposure to radiation dosage.
  • The diagnosis time is shorter and at lower cost.