Clinical Trial Validation of Automated Segmentation and Scoring of Pulmonary Cysts in Thoracic CT Scans.

In cystic lung diseases such as lymphangioleiomyomatosis (LAM), a CT-based cyst score that measures the percentage of the lung volume occupied by cysts is a common index of the cyst burden in the lungs. Although the current semi-automatic measurement of the cyst score is well established, it is susceptible to human operator variabilities. We recently developed a fully automatic method incorporating adaptive features in place of manual adjustments. In this clinical study, the automatic method is validated against the standard method in several aspects. These include the agreement between the cyst scores of the two methods, the agreement of each method with independent tests of pulmonary function, and the temporal consistency of the measurements in the consecutive visits of the same patients. We found that the automatic method agreed with the standard method as well as the agreement between two trained operators running the same standard method; both methods obtained the same level of correlation with laboratory pulmonary function tests; the automated method had better temporal consistency than the standard method (p < 0.0001). The study indicates that the automatic method could replace the standard method and provide better consistency in assessing the extent of cystic changes in the lungs of patients.

Automated Segmentation and Measurements of Pulmonary Cysts in Lymphangioleiomyomatosis across Multiple CT Scanner Platforms over a Period of Two Decades.

(1) Background: Lymphangioleiomyomatosis is a genetic disease that affects mostly women of childbearing age. In the lungs, it manifests as the progressive formation of air-filled cysts and is associated with a decline in lung function. With a median survival of 29 years after the onset of symptoms, computed-tomographic monitoring of cystic changes in the lungs is a key part of the management of the disease. However, the current standard method to measure cyst burdens from CT is semi-automatic and requires manual adjustments from trained operators to obtain consistent results due to variabilities in CT technology and imaging conditions over the long course of the disease. This can be impractical for longitudinal studies involving large numbers of scans and is susceptible to subjective biases. (2) Methods: We developed an automated method of pulmonary cyst segmentation for chest CT images incorporating novel graphics processing algorithms. We assessed its performance against the gold-standard semi-automated method performed by experienced operators who were blinded to the results of the automated method. (3) Results: the automated method had the same consistency over time as the gold-standard method, but its cyst scores were more strongly correlated with concurrent pulmonary function results from the physiology laboratory than those of the gold-standard method. (4) Conclusions: The automated cyst segmentation is a competent replacement for the gold-standard semi-automated process. It is a solution for saving time and labor in clinical studies of lymphangioleiomyomatosis that may involve large numbers of chest CT scans from diverse scanner platforms and protocols.