MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Deriving Ventilation Imaging From 4DCT by Deep Convolutional Neural Network

Y Zhong1*, Y Vinogradskiy2 , L Chen1 , N Myziuk3 , R Castillo4 , E Castillo3 , T Guerrero3 , S Jiang1 , J Wang1 , (1) UT Southwestern Medical Center, Dallas, TX, (2) University of Colorado Denver, Aurora, CO, (3) Beaumont Health System, Royal Oak, MI, (4) Emory Univ, Atlanta, GA

Presentations

(Sunday, 7/29/2018) 2:05 PM - 3:00 PM

Room: Room 207

Purpose: Ventilation images can be derived from 4DCT through analyzing the change of HU values and deformable vector fields between different respiration phases of CT. As deformable image registration (DIR) is involved, accuracy of 4DCT derived ventilation image is sensitive to the choice of DIR algorithms. To overcome the uncertainty associated with DIR, we develop a method based on deep convolutional neural network (CNN) to derive ventilation images directly from the 4DCT without explicit image registration.

Methods: 4DCT and ventilation images from 82 lung cancer patients were used in this study. The training dataset consisted of paired ventilation and 4DCT images from 75 patients. The independent testing dataset consisted of 4DCT images from seven patients. In the proposed CNN architecture, the CT two-channel input data consists of CT at the end of exhale and the end of inhale phases. The first convolutional layer has 32 different kernels of size 5×5×5, followed by another 8 convolutional layers each of which is equipped with an activation layer (ReLU). The loss function is the mean-squared-error (MSE) to measure the intensity difference between the predicted and reference ventilation images. The ventilation images calculated using the CNN method were compared against ventilation images calculated with the standard DIR/HU change algorithm.

Results: The features on the predicted ventilation images of the test patients resembled the reference images. The structure similarity index (SSIM) and the correlation coefficients averaged over the test cases were 0.928±0.056 and 0.939±0.047, respectively, indicating a high degree of similarity between the predicted and reference ventilation images.

Conclusion: Our preliminary results demonstrate that the proposed CNN-based method is able to generate ventilation images with high fidelity compared to the reference images. By avoiding explicit DIR, CNN-based method developed here has potential to improve consistency of ventilation images derived from 4DCT.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by US National Institutes of Health (R01 EB020366).

Keywords

Ventilation/perfusion, CT, Functional Imaging

Taxonomy

IM- Dataset analysis/biomathematics: Machine learning

Contact Email