Click here to


Are you sure ?

Yes, do it No, cancel

Clinical Validation of Four-Dimensional Computed Tomography-Based Lung Ventilation Images Using Texture Feature Analysis and Pulmonary Function Test

H Zhang1*, M Huang2 , S Feigenberg2 , A Berman2 , W Lu3 , V Jain2 , Y Xiao2 , P Jin2 , P Mohindra1 , C Simone1 , S Bentzen1 , W D'Souza1 , (1) University of Maryland School of Medicine, Baltimore, MD, (2) University of Pennsylvania, Philadelphia, PA, (3) Memorial Sloan Kettering Cancer Center, New York, NY,


(Wednesday, 8/1/2018) 7:30 AM - 9:30 AM

Room: Karl Dean Ballroom C

Purpose: To validate four-dimensional computed tomography (4DCT)-based lung ventilation images using texture feature analysis and pulmonary function test (PFT).

Methods: 48 consecutive locally advanced non-small cell lung cancer patients who had pre-therapy PFTs and 4DCTs were included. Lung ventilation images were obtained after registering the end-of-inhalation phase to the end-of-exhalation phase of the 4DCT and calculating the Hounsfield units (HU) change, corrected with the change of mass over the respiratory cycle. Tidal volume (TV) was calculated from the derived ventilation images. As ventilation image heterogeneity that can be identified from texture analysis is related to lung function, we extracted nine texture features from the ventilation image, including energy, contrast, entropy, homogeneity, correlation, SumAverage, variance, dissimilarity and AutoCorrelation. These features were correlated with PFT measurements: forced expiratory volume measured in 1 sec (FEV1) and diffusing capacity of carbon monoxide (DLCO). Standard threshold of 80% for DLCO was used to define normal versus abnormal lung function. Univariate and multivariate logistic regression analyses were conducted to examine the features’ ability to predict lung function.

Results: TV was correlated with DLCO (correlation coefficient of 0.32, p = 0.046). Among texture features, correlation, which characterizes the consistency of ventilation texture, correlated with FEV1 (correlation coefficient of 0.32, p = 0.029), but appeared to be more correlated with DLCO (correlation coefficient of 0.52, p < 0.001). Single feature, correlation, had the highest prediction accuracy of lung function (AUC of 0.76). On multivariate analysis, features including energy, entropy, correlation, SumAverage, variance, dissimilarity and AutoCorrelation achieved an AUC of 0.88 when predicting lung function.

Conclusion: We have shown that features extracted from 4DCT-based ventilation images are predictive of global lung function measured by PFTs, making it feasible to incorporate these images into radiotherapy treatment planning to more optimally preserve lung function than current planning allows.


Ventilation/perfusion, Texture Analysis, Functional Imaging



Contact Email